classifier_species_expt3.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training species classifier Expt 3: cross-validation of Oswald data\n",
    "## Trained on all except the select data\n",
    "## Four datasets: Oswald, Gillispie, DCLDE 2011 & Watkin\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from itertools import permutations\n",
    "import random\n",
    "import os\n",
    "import glob\n",
    "import pickle\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.patheffects as PathEffects\n",
    "import pandas as pd\n",
    "from os import makedirs\n",
    "from datetime import datetime\n",
    "\n",
    "from math import floor\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.utils import shuffle\n",
    "from sklearn.utils.class_weight import compute_class_weight\n",
    "from sklearn.metrics import roc_curve\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.metrics import confusion_matrix, classification_report\n",
    "\n",
    "# from MulticoreTSNE import MulticoreTSNE as TSNE\n",
    "# from sklearn.manifold import TSNE\n",
    "import seaborn as sns\n",
    "\n",
    "from tensorflow.keras.utils import Sequence\n",
    "from tensorflow.keras.layers import Input, Dense, BatchNormalization, Dropout, Activation\n",
    "from tensorflow.keras.layers import Conv2D, Lambda, Flatten, MaxPooling2D, Concatenate, LSTM, Reshape, Lambda, ConvLSTM2D, GlobalAveragePooling2D, GlobalMaxPooling2D\n",
    "from tensorflow.keras.layers import Bidirectional, LSTM, GRU\n",
    "from tensorflow.keras.models import Model, Sequential\n",
    "from tensorflow.keras.regularizers import l2\n",
    "from tensorflow.keras import backend as K\n",
    "\n",
    "from tensorflow.keras.optimizers.schedules import ExponentialDecay\n",
    "from tensorflow.keras.optimizers import SGD,Adam\n",
    "from tensorflow.keras.losses import binary_crossentropy, CategoricalCrossentropy\n",
    "from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard, EarlyStopping\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "from tensorflow.keras.models import load_model\n",
    "# import tensorflow_addons as tfa\n",
    "# import tensorflow_datasets as tfds\n",
    "from tensorflow.math import l2_normalize\n",
    "\n",
    "from tensorflow.keras.optimizers.schedules import PiecewiseConstantDecay\n",
    "\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\", category=DeprecationWarning)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Hyper parameters\n",
    "learning_rate = 1.0e-4\n",
    "conv_dim = 16\n",
    "rnn_dim = 16\n",
    "pool_size = 2\n",
    "pool_stride = 2\n",
    "l2_regu = 0.000\n",
    "drop_rate = 0.2\n",
    "hidden_units = 512\n",
    "fcn_dim = 512\n",
    "\n",
    "# learning_rate = 1.e-4\n",
    "# conv_dim = 64\n",
    "# rnn_dim = 16\n",
    "# pool_size = 2\n",
    "# pool_stride = 2\n",
    "# l2_regu = 0.00\n",
    "# drop_rate = 0.2\n",
    "# # drop_rate = 0.5\n",
    "# hidden_units = 512\n",
    "# fcn_dim = 512\n",
    "\n",
    "num_epoch = 200\n",
    "# batch_size = 128\n",
    "batch_size = 32  # for cnn14+attention\n",
    "\n",
    "num_species = 21\n",
    "num_patience = 20"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# data_type_dict = {1: 'universal', 2: 'file', 3: 'encounter', 4: 'domain'}\n",
    "# data_type = 2\n",
    "\n",
    "work_path = '/home/ys587/__Data/__whistle/__whislte_30_species'\n",
    "fit_result_path =  os.path.join(work_path, '__fit_result_species')\n",
    "# feature_path = os.path.join(work_path, '__feature_species')\n",
    "feature_path = os.path.join(work_path, '__dataset/20210210')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# species_dict = {'BD': 0, 'MH': 1, 'CD': 2, 'STR': 3, 'SPT': 4, 'SPIN': 5, 'PLT': 6, 'RD': 7, 'RT': 8,\n",
    "#                 'WSD': 9, 'FKW': 10, 'BEL': 11, 'KW': 12, 'WBD': 13, 'DUSK': 14, 'FRA': 15, 'PKW': 16, 'LPLT': 17, }\n",
    "species_dict = {'BD': 0, 'MH': 1, 'CD': 2, 'STR': 3, 'SPT': 4, 'SPIN': 5, 'PLT': 6, 'RD': 7, 'RT': 8,\n",
    "                'WSD': 9, 'FKW': 10, 'BEL': 11, 'KW': 12, 'WBD': 13, 'DUSK': 14, 'FRA': 15, 'PKW': 16, 'LPLT': 17,\n",
    "                'CLY': 18, 'SPE': 19, 'ASP': 20}\n",
    "species_list = list(species_dict.keys())\n",
    "species_id = list(species_dict.values())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Training data\n",
    "fea_temp = np.load(os.path.join(feature_path, 'train_no_oswald_STAR2000.npz'))\n",
    "fea_train = fea_temp['fea_train']\n",
    "label_train_list = fea_temp['label_train']\n",
    "del fea_temp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_train = np.zeros(len(label_train_list))\n",
    "for ii in range(len(label_train_list)):\n",
    "    label_train[ii] = species_dict[label_train_list[ii]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Testing data\n",
    "fea_temp = np.load(os.path.join(feature_path, 'oswald_STAR2000_orig.npz'))\n",
    "fea_test = fea_temp['feas_orig']\n",
    "label_test_list = fea_temp['labels_orig']\n",
    "\n",
    "fea_test = fea_test[:,:100,:]\n",
    "label_test = np.zeros(len(label_test_list))\n",
    "for ii in range(len(label_test_list)):\n",
    "    label_test[ii] = species_dict[label_test_list[ii]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "fea_train = fea_train[:,:100,:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_train = np.zeros(len(label_train_list))\n",
    "for ii in range(len(label_train_list)):\n",
    "    label_train[ii] = species_dict[label_train_list[ii]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({'BD': 40648,\n",
       "         'CD': 31972,\n",
       "         'FKW': 22152,\n",
       "         'SPIN': 15240,\n",
       "         'SPT': 21556,\n",
       "         'STR': 17432,\n",
       "         'PLT': 17280,\n",
       "         'RT': 4980,\n",
       "         'BEL': 1168,\n",
       "         'DUSK': 2052,\n",
       "         'FRA': 3288,\n",
       "         'KW': 8064,\n",
       "         'LPLT': 4320,\n",
       "         'PKW': 228,\n",
       "         'RD': 7412,\n",
       "         'WSD': 16676,\n",
       "         'WBD': 1608,\n",
       "         'MH': 82272,\n",
       "         'ASP': 360,\n",
       "         'CLY': 904,\n",
       "         'SPE': 27028})"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from collections import Counter\n",
    "Counter(label_train_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Counter({0.0: 121, 2.0: 3964, 6.0: 31, 8.0: 76, 5.0: 491, 4.0: 845, 3.0: 1140})"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Counter(label_test.tolist())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create a folder based on date & time\n",
    "today = datetime.now()\n",
    "\n",
    "fit_result_path1 = os.path.join(fit_result_path, today.strftime('%Y%m%d_%H%M%S'))\n",
    "\n",
    "if not os.path.exists(fit_result_path1):\n",
    "    makedirs(fit_result_path1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "feature train shape: (326640, 100, 128)\n",
      "feature test shape: (6668, 100, 128)\n",
      "label train shape: (326640,)\n",
      "label test shape: (6668,)\n",
      "dim_time: 100\n",
      "dim_freq: 128\n"
     ]
    }
   ],
   "source": [
    "print('feature train shape: '+str(fea_train.shape))\n",
    "print('feature test shape: '+str(fea_test.shape))\n",
    "print('label train shape: '+str(label_train.shape))\n",
    "print('label test shape: '+str(label_test.shape))\n",
    "\n",
    "dim_time = fea_train.shape[1]\n",
    "dim_freq = fea_train.shape[2]\n",
    "print('dim_time: '+str(dim_time))\n",
    "print('dim_freq: '+str(dim_freq))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# shuffle features & labels\n",
    "fea_train, label_train = shuffle(fea_train, label_train, random_state=0)\n",
    "fea_test, label_test = shuffle(fea_test, label_test, random_state=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# data generator\n",
    "class DataGenerator(tf.keras.utils.Sequence):\n",
    "    def __init__(self, feature, label, batch_size=32, num_classes=None, shuffle=True):\n",
    "        self.batch_size = batch_size\n",
    "        self.X = feature\n",
    "        self.X_dim = len(feature.shape)\n",
    "        self.y = to_categorical(label, num_classes)\n",
    "        self.indices = np.arange(self.y.shape[0])\n",
    "        self.num_classes = num_classes\n",
    "        self.shuffle = shuffle\n",
    "        \n",
    "        # self.index = np.arange(len(self.indices))\n",
    "        #self.df = dataframe\n",
    "        #self.indices = self.df.index.tolist()        \n",
    "        # self.x_col = x_col\n",
    "        # self.y_col = y_col\n",
    "        \n",
    "        self.on_epoch_end()\n",
    "\n",
    "    def __len__(self):\n",
    "        return int(floor(len(self.indices)/self.batch_size))\n",
    "        # return label.shape[0]\n",
    "\n",
    "    def __getitem__(self, index):\n",
    "        # index = self.index[index * self.batch_size:(index + 1) * self.batch_size]\n",
    "        # batch = [self.indices[k] for k in index]\n",
    "        batch = list(range(index*self.batch_size, (index+1)*self.batch_size))\n",
    "        \n",
    "        X, y = self.__get_data(batch)\n",
    "        return X, y\n",
    "\n",
    "    def on_epoch_end(self):\n",
    "        if self.shuffle == True:\n",
    "            np.random.shuffle(self.indices)\n",
    "\n",
    "    def __get_data(self, batch):\n",
    "        y = np.zeros((self.batch_size, self.y.shape[1]))\n",
    "        \n",
    "        if self.X_dim == 3:\n",
    "            X = np.zeros((self.batch_size, self.X.shape[1], self.X.shape[2]))\n",
    "            for i, id in enumerate(batch):\n",
    "                X[i,:, :] = self.X[id, :, :]  # logic\n",
    "                y[i, :] = self.y[id, :] # labels\n",
    "                \n",
    "        elif self.X_dim == 4:\n",
    "            X = np.zeros((self.batch_size, self.X.shape[1], self.X.shape[2], self.X.shape[3]))\n",
    "            for i, id in enumerate(batch):\n",
    "                X[i,:, :, :] = self.X[id, :, :, :]  # logic\n",
    "                y[i, :] = self.y[id, :] # labels\n",
    "\n",
    "        return X, y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "fea_train = np.expand_dims(fea_train, axis=3)\n",
    "fea_test = np.expand_dims(fea_test, axis=3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "fea_train, fea_validate, label_train, label_validate = train_test_split(fea_train, label_train, test_size=0.30, random_state=42)\n",
    "\n",
    "train_generator = DataGenerator(fea_train, label_train, batch_size=batch_size, num_classes=num_species)\n",
    "del fea_train\n",
    "validate_generator = DataGenerator(fea_validate, label_validate, batch_size=batch_size, num_classes=num_species)\n",
    "del fea_validate\n",
    "\n",
    "# test_generator = DataGenerator(fea_test, label_test, batch_size=batch_size, num_classes=num_species)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Kong's attention\n",
    "# def max_pooling(inputs, **kwargs):\n",
    "#     input = inputs[0]   # (batch_size, time_steps, freq_bins)\n",
    "#     return K.max(input, axis=1)\n",
    "def max_pooling(inputs, **kwargs):\n",
    "    # input = inputs[0]   # (batch_size, time_steps, freq_bins)\n",
    "    return K.max(inputs, axis=1)\n",
    "\n",
    "\n",
    "def average_pooling(inputs, **kwargs):\n",
    "    input = inputs[0]   # (batch_size, time_steps, freq_bins)\n",
    "    return K.mean(input, axis=1)\n",
    "\n",
    "\n",
    "def attention_pooling(inputs, **kwargs):\n",
    "    [out, att] = inputs\n",
    "\n",
    "    epsilon = 1e-7\n",
    "    att = K.clip(att, epsilon, 1. - epsilon)\n",
    "    normalized_att = att / K.sum(att, axis=1)[:, None, :]\n",
    "\n",
    "    return K.sum(out * normalized_att, axis=1)\n",
    "\n",
    "\n",
    "def pooling_shape(input_shape):\n",
    "\n",
    "    if isinstance(input_shape, list):\n",
    "        (sample_num, time_steps, freq_bins) = input_shape[0]\n",
    "\n",
    "    else:\n",
    "        (sample_num, time_steps, freq_bins) = input_shape\n",
    "\n",
    "    return (sample_num, freq_bins)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# cnn14 attention with customized maxpooling\n",
    "def model_cnn14_attention_multi(time_steps, freq_bins, classes_num, model_type='feature_level_attention', conv_dim=64, rnn_dim=128, pool_size=2, pool_stride=2, hidden_units=512, l2_regu=0., drop_rate=0., multilabel=True):\n",
    "    # Kong's attention\n",
    "    # model_type = 'decision_level_max_pooling'  # problem with dimensions of the Lambda layer after training\n",
    "    # model_type = 'decision_level_average_pooling' # problem with dimensions of the Lambda layer after training\n",
    "    # model_type = 'decision_level_single_attention'\n",
    "    # model_type = 'decision_level_multi_attention'\n",
    "    # model_type = 'feature_level_attention'\n",
    "\n",
    "    input_layer = Input(shape=(time_steps, freq_bins, 1), name='input')\n",
    "    # group 1\n",
    "    y = Conv2D(conv_dim, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(input_layer)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 2\n",
    "    y = Conv2D(conv_dim*2, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*2, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 3\n",
    "    y = Conv2D(conv_dim*4, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*4, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 4 \n",
    "    y = Conv2D(conv_dim*8, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*8, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "\n",
    "    # group 5\n",
    "    y = Conv2D(conv_dim*16, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*16, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(1, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "\n",
    "    # group 6\n",
    "    y = Conv2D(conv_dim*32, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*32, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(1, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # change dimensions: samples, time, frequency, channels => samples, time, frequency*channels\n",
    "    dim_cnn = K.int_shape(y)\n",
    "    y = Reshape((dim_cnn[1], dim_cnn[2]*dim_cnn[3]))(y)\n",
    "\n",
    "    a1 = Dense(hidden_units)(y)\n",
    "    a1 = BatchNormalization()(a1)\n",
    "    a1 = Activation('relu')(a1)\n",
    "    a1 = Dropout(drop_rate)(a1)\n",
    "\n",
    "    a2 = Dense(hidden_units)(a1)\n",
    "    a2 = BatchNormalization()(a2)\n",
    "    a2 = Activation('relu')(a2)\n",
    "    a2 = Dropout(drop_rate)(a2)\n",
    "\n",
    "    a3 = Dense(hidden_units)(a2)\n",
    "    a3 = BatchNormalization()(a3)\n",
    "    a3 = Activation('relu')(a3)\n",
    "    a3 = Dropout(drop_rate)(a3)\n",
    "\n",
    "    # Pooling layers\n",
    "    if model_type == 'decision_level_max_pooling':\n",
    "        '''Global max pooling.\n",
    "\n",
    "        [1] Choi, Keunwoo, et al. \"Automatic tagging using deep convolutional \n",
    "        neural networks.\" arXiv preprint arXiv:1606.00298 (2016).\n",
    "        '''\n",
    "        cla = Dense(classes_num, activation='sigmoid')(a3)\n",
    "        \n",
    "        # output_layer = Lambda(\n",
    "        #    max_pooling, \n",
    "        #    output_shape=pooling_shape)(\n",
    "        #    [cla])\n",
    "        output_layer = Lambda(max_pooling)(cla)\n",
    "\n",
    "    elif model_type == 'decision_level_average_pooling':\n",
    "        '''Global average pooling.\n",
    "\n",
    "        [2] Lin, Min, et al. Qiang Chen, and Shuicheng Yan. \"Network in \n",
    "        network.\" arXiv preprint arXiv:1312.4400 (2013).\n",
    "        '''\n",
    "        cla = Dense(classes_num, activation='sigmoid')(a3)\n",
    "        # output_layer = Lambda(\n",
    "        #    average_pooling,\n",
    "        #    output_shape=pooling_shape)(\n",
    "        #    [cla])\n",
    "        output_layer = Lambda(average_pooling)(cla)\n",
    "\n",
    "    elif model_type == 'decision_level_single_attention':\n",
    "        '''Decision level single attention pooling.\n",
    "        [3] Kong, Qiuqiang, et al. \"Audio Set classification with attention\n",
    "        model: A probabilistic perspective.\" arXiv preprint arXiv:1711.00927\n",
    "        (2017).\n",
    "        '''\n",
    "        cla = Dense(classes_num, activation='sigmoid')(a3)\n",
    "        att = Dense(classes_num, activation='softmax')(a3)\n",
    "        output_layer = Lambda(\n",
    "            attention_pooling, output_shape=pooling_shape)([cla, att])\n",
    "\n",
    "    elif model_type == 'decision_level_multi_attention':\n",
    "        '''Decision level multi attention pooling.\n",
    "        [4] Yu, Changsong, et al. \"Multi-level Attention Model for Weakly\n",
    "        Supervised Audio Classification.\" arXiv preprint arXiv:1803.02353\n",
    "        (2018).\n",
    "        '''\n",
    "        cla1 = Dense(classes_num, activation='sigmoid')(a2)\n",
    "        att1 = Dense(classes_num, activation='softmax')(a2)\n",
    "        out1 = Lambda(\n",
    "            attention_pooling, output_shape=pooling_shape)([cla1, att1])\n",
    "\n",
    "        cla2 = Dense(classes_num, activation='sigmoid')(a3)\n",
    "        att2 = Dense(classes_num, activation='softmax')(a3)\n",
    "        out2 = Lambda(\n",
    "            attention_pooling, output_shape=pooling_shape)([cla2, att2])\n",
    "\n",
    "        b1 = Concatenate(axis=-1)([out1, out2])\n",
    "        b1 = Dense(classes_num)(b1)\n",
    "        \n",
    "        if multilabel:\n",
    "            output_layer = Activation('sigmoid')(b1)\n",
    "        else:\n",
    "            output_layer = Activation('softmax')(b1)\n",
    "\n",
    "    elif model_type == 'feature_level_attention':\n",
    "        '''Feature level attention.\n",
    "        [1] Kong, Qiuqiang, et al. \"Weakly labelled audioset tagging with \n",
    "        attention neural networks.\" (2019).\n",
    "        '''\n",
    "        cla = Dense(hidden_units, activation='linear')(a3)\n",
    "        att = Dense(hidden_units, activation='sigmoid')(a3)\n",
    "        b1 = Lambda(\n",
    "            attention_pooling, output_shape=pooling_shape)([cla, att])\n",
    "\n",
    "        b1 = BatchNormalization()(b1)\n",
    "        b1 = Activation(activation='relu')(b1)\n",
    "        b1 = Dropout(drop_rate)(b1)\n",
    "        \n",
    "        if multilabel:\n",
    "            output_layer = Dense(classes_num, activation='sigmoid')(b1)\n",
    "        else:\n",
    "            output_layer = Dense(classes_num, activation='softmax')(b1)\n",
    "\n",
    "    else:\n",
    "        raise Exception(\"Incorrect model_type!\")\n",
    "\n",
    "    # Build model\n",
    "    model = Model(inputs=input_layer, outputs=output_layer)\n",
    "    \n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# cnn14 attention\n",
    "def model_cnn14_attention(time_steps, freq_bins, classes_num, model_type='feature_level_attention', conv_dim=64, rnn_dim=128, pool_size=2, pool_stride=2, hidden_units=512, l2_regu=0., drop_rate=0., multilabel=True):\n",
    "    # Kong's attention\n",
    "    # model_type = 'decision_level_max_pooling'  # problem with dimensions of the Lambda layer after training\n",
    "    # model_type = 'decision_level_average_pooling' # problem with dimensions of the Lambda layer after training\n",
    "    # model_type = 'decision_level_single_attention'\n",
    "    # model_type = 'decision_level_multi_attention'\n",
    "    # model_type = 'feature_level_attention'\n",
    "\n",
    "    input_layer = Input(shape=(time_steps, freq_bins, 1), name='input')\n",
    "    # group 1\n",
    "    y = Conv2D(conv_dim, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(input_layer)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 2\n",
    "    y = Conv2D(conv_dim*2, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*2, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 3\n",
    "    y = Conv2D(conv_dim*4, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*4, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 4 \n",
    "    y = Conv2D(conv_dim*8, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*8, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "\n",
    "    # group 5\n",
    "    y = Conv2D(conv_dim*16, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*16, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "\n",
    "    # group 6\n",
    "    y = Conv2D(conv_dim*32, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*32, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(pool_size, 2), strides=(pool_stride, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # change dimensions: samples, time, frequency, channels => samples, time, frequency*channels\n",
    "    dim_cnn = K.int_shape(y)\n",
    "    y = Reshape((dim_cnn[1], dim_cnn[2]*dim_cnn[3]))(y)\n",
    "\n",
    "    a1 = Dense(hidden_units)(y)\n",
    "    a1 = BatchNormalization()(a1)\n",
    "    a1 = Activation('relu')(a1)\n",
    "    a1 = Dropout(drop_rate)(a1)\n",
    "\n",
    "    a2 = Dense(hidden_units)(a1)\n",
    "    a2 = BatchNormalization()(a2)\n",
    "    a2 = Activation('relu')(a2)\n",
    "    a2 = Dropout(drop_rate)(a2)\n",
    "\n",
    "    a3 = Dense(hidden_units)(a2)\n",
    "    a3 = BatchNormalization()(a3)\n",
    "    a3 = Activation('relu')(a3)\n",
    "    a3 = Dropout(drop_rate)(a3)\n",
    "\n",
    "    # Pooling layers\n",
    "    if model_type == 'decision_level_max_pooling':\n",
    "        '''Global max pooling.\n",
    "\n",
    "        [1] Choi, Keunwoo, et al. \"Automatic tagging using deep convolutional \n",
    "        neural networks.\" arXiv preprint arXiv:1606.00298 (2016).\n",
    "        '''\n",
    "        cla = Dense(classes_num, activation='sigmoid')(a3)\n",
    "        \n",
    "        # output_layer = Lambda(\n",
    "        #    max_pooling, \n",
    "        #    output_shape=pooling_shape)(\n",
    "        #    [cla])\n",
    "        output_layer = Lambda(max_pooling)(cla)\n",
    "\n",
    "    elif model_type == 'decision_level_average_pooling':\n",
    "        '''Global average pooling.\n",
    "\n",
    "        [2] Lin, Min, et al. Qiang Chen, and Shuicheng Yan. \"Network in \n",
    "        network.\" arXiv preprint arXiv:1312.4400 (2013).\n",
    "        '''\n",
    "        cla = Dense(classes_num, activation='sigmoid')(a3)\n",
    "        # output_layer = Lambda(\n",
    "        #    average_pooling,\n",
    "        #    output_shape=pooling_shape)(\n",
    "        #    [cla])\n",
    "        output_layer = Lambda(average_pooling)(cla)\n",
    "\n",
    "    elif model_type == 'decision_level_single_attention':\n",
    "        '''Decision level single attention pooling.\n",
    "        [3] Kong, Qiuqiang, et al. \"Audio Set classification with attention\n",
    "        model: A probabilistic perspective.\" arXiv preprint arXiv:1711.00927\n",
    "        (2017).\n",
    "        '''\n",
    "        cla = Dense(classes_num, activation='sigmoid')(a3)\n",
    "        att = Dense(classes_num, activation='softmax')(a3)\n",
    "        output_layer = Lambda(\n",
    "            attention_pooling, output_shape=pooling_shape)([cla, att])\n",
    "\n",
    "    elif model_type == 'decision_level_multi_attention':\n",
    "        '''Decision level multi attention pooling.\n",
    "        [4] Yu, Changsong, et al. \"Multi-level Attention Model for Weakly\n",
    "        Supervised Audio Classification.\" arXiv preprint arXiv:1803.02353\n",
    "        (2018).\n",
    "        '''\n",
    "        cla1 = Dense(classes_num, activation='sigmoid')(a2)\n",
    "        att1 = Dense(classes_num, activation='softmax')(a2)\n",
    "        out1 = Lambda(\n",
    "            attention_pooling, output_shape=pooling_shape)([cla1, att1])\n",
    "\n",
    "        cla2 = Dense(classes_num, activation='sigmoid')(a3)\n",
    "        att2 = Dense(classes_num, activation='softmax')(a3)\n",
    "        out2 = Lambda(\n",
    "            attention_pooling, output_shape=pooling_shape)([cla2, att2])\n",
    "\n",
    "        b1 = Concatenate(axis=-1)([out1, out2])\n",
    "        b1 = Dense(classes_num)(b1)\n",
    "        \n",
    "        if multilabel:\n",
    "            output_layer = Activation('sigmoid')(b1)\n",
    "        else:\n",
    "            output_layer = Activation('softmax')(b1)\n",
    "\n",
    "    elif model_type == 'feature_level_attention':\n",
    "        '''Feature level attention.\n",
    "        [1] Kong, Qiuqiang, et al. \"Weakly labelled audioset tagging with \n",
    "        attention neural networks.\" (2019).\n",
    "        '''\n",
    "        cla = Dense(hidden_units, activation='linear')(a3)\n",
    "        att = Dense(hidden_units, activation='sigmoid')(a3)\n",
    "        b1 = Lambda(\n",
    "            attention_pooling, output_shape=pooling_shape)([cla, att])\n",
    "\n",
    "        b1 = BatchNormalization()(b1)\n",
    "        b1 = Activation(activation='relu')(b1)\n",
    "        b1 = Dropout(drop_rate)(b1)\n",
    "        \n",
    "        if multilabel:\n",
    "            output_layer = Dense(classes_num, activation='sigmoid')(b1)\n",
    "        else:\n",
    "            output_layer = Dense(classes_num, activation='softmax')(b1)\n",
    "\n",
    "    else:\n",
    "        raise Exception(\"Incorrect model_type!\")\n",
    "\n",
    "    # Build model\n",
    "    model = Model(inputs=input_layer, outputs=output_layer)\n",
    "    \n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "def model_cnn14(time_steps, freq_bins, classes_num, pool_size=2, pool_stride=2, conv_dim=16, fcn_dim=512, l2_regu=0., drop_rate=0.):\n",
    "    # input\n",
    "    input_layer = Input(shape=(time_steps, freq_bins, 1), name='input')\n",
    "    \n",
    "    # group 1\n",
    "    y = Conv2D(conv_dim, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(input_layer)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 2\n",
    "    y = Conv2D(conv_dim*2, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*2, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 3\n",
    "    y = Conv2D(conv_dim*4, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*4, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "    \n",
    "    # group 4 \n",
    "    y = Conv2D(conv_dim*8, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*8, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "\n",
    "    # group 5\n",
    "    y = Conv2D(conv_dim*16, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*16, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same')(y)\n",
    "    y = Dropout(drop_rate)(y)\n",
    "\n",
    "    # group 6\n",
    "    y = Conv2D(conv_dim*32, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    y = Conv2D(conv_dim*32, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(l2_regu))(y)\n",
    "    y = BatchNormalization()(y)\n",
    "    y = Activation(activation='relu')(y)\n",
    "    \n",
    "    y = GlobalMaxPooling2D()(y)\n",
    "    \n",
    "    # FC block\n",
    "    y = Dense(fcn_dim, activation='relu', name='cnn14_fcn')(y)  # original 512\n",
    "    x = Dense(classes_num, activation='softmax')(y)\n",
    "    \n",
    "    # Build model\n",
    "    model = Model(inputs=input_layer, outputs=x)\n",
    "    \n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Model compile, class weight & fitting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# def find_best_model(model_folder, remove_others=False):\n",
    "#     model_list = glob.glob(model_folder+'/*.hdf5')\n",
    "#     model_list.sort()\n",
    "#     the_best_model = model_list[-1]\n",
    "    \n",
    "#     if remove_others==True:\n",
    "#         for mm in model_list[:-1]:\n",
    "#             os.remove(mm)\n",
    "            \n",
    "#     print(the_best_model)\n",
    "    \n",
    "#     return the_best_model\n",
    "import re\n",
    "\n",
    "def find_best_model(classifier_path, fmt='epoch_\\d+_valloss_(\\d+.\\d{4})_valacc_\\d+.\\d{4}.hdf5', is_max=False, purge=True):\n",
    "    \"\"\"\n",
    "    Return the path to the model with the best accuracy, given the path to\n",
    "    all the trained classifiers\n",
    "    Args:\n",
    "        classifier_path: path to all the trained classifiers\n",
    "        fmt: e.g. \"epoch_\\d+_[0-1].\\d+_(\\d+.\\d{4}).hdf5\"\n",
    "        'epoch_\\d+_valloss_(\\d+.\\d{4})_valacc_\\d+.\\d{4}.hdf5'\n",
    "        is_max: use max; otherwise, min\n",
    "        purge: True to purge models files except the best one\n",
    "    Return:\n",
    "        the path of the model with the best accuracy\n",
    "    \"\"\"\n",
    "    # list all files ending with .hdf5\n",
    "    day_list = sorted(glob.glob(os.path.join(classifier_path + '/', '*.hdf5')))\n",
    "\n",
    "    # re the last 4 digits for accuracy\n",
    "    hdf5_filename = []\n",
    "    hdf5_accu = np.zeros(len(day_list))\n",
    "    for dd in range(len(day_list)):\n",
    "        filename = os.path.basename(day_list[dd])\n",
    "        hdf5_filename.append(filename)\n",
    "        # m = re.search(\"_F1_(0.\\d{4}).hdf5\", filename)\n",
    "        # m = re.search(\"_([0-1].\\d{4}).hdf5\", filename)\n",
    "        # m = re.search(\"epoch_\\d+_[0-1].\\d+_(\\d+.\\d{4}).hdf5\", filename)\n",
    "        m = re.search(fmt, filename)\n",
    "        try:\n",
    "            hdf5_accu[dd] = float(m.groups()[0])\n",
    "        except:\n",
    "            continue\n",
    "\n",
    "    # select the laregest one and write to the variable classifier_file\n",
    "    if len(hdf5_accu) == 0:\n",
    "        best_model_path = ''\n",
    "        best_accu = 0\n",
    "    else:\n",
    "        if is_max is True:\n",
    "            ind_max = np.argmax(hdf5_accu)\n",
    "        else: # use min instead\n",
    "            ind_max = np.argmin(hdf5_accu)\n",
    "        best_model_path = day_list[int(ind_max)]\n",
    "        best_accu = hdf5_accu[ind_max]\n",
    "        # purge all model files except the best_model\n",
    "        if purge:\n",
    "            for ff in day_list:\n",
    "                if ff != best_model_path:\n",
    "                    os.remove(ff)\n",
    "    print('Best model:'+str(best_accu))\n",
    "    print(best_model_path)\n",
    "    return best_model_path, best_accu\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "### cnn4 + attention\n",
    "# model = model_cnn_attention(dim_time, dim_freq, num_species, model_type='decision_level_max_pooling', conv_dim=conv_dim, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "# model = model_cnn_attention(dim_time, dim_freq, num_species, model_type='decision_level_multi_attention', conv_dim=conv_dim, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "# model = model_cnn_attention(dim_time, dim_freq, num_species, model_type='feature_level_attention', conv_dim=conv_dim, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "\n",
    "# vggish\n",
    "# model = model_vggish(dim_time, dim_freq, num_species, conv_dim=conv_dim, fcn_dim=fcn_dim)\n",
    "\n",
    "# cnn10\n",
    "# model = model_cnn10(dim_time, dim_freq, num_species, conv_dim=conv_dim, fcn_dim=fcn_dim, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "\n",
    "# cnn14\n",
    "# model = model_cnn14(dim_time, dim_freq, num_species, conv_dim=conv_dim, fcn_dim=fcn_dim, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "\n",
    "# cnn14 attention\n",
    "# model = model_cnn14_attention(dim_time, dim_freq, num_species, model_type='feature_level_attention', conv_dim=conv_dim, pool_size=pool_size, pool_stride=pool_stride, hidden_units=hidden_units, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "model = model_cnn14_attention_multi(dim_time, dim_freq, num_species, model_type='feature_level_attention', conv_dim=conv_dim, pool_size=pool_size, pool_stride=pool_stride, hidden_units=hidden_units, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "# model = model_cnn14_bigru_attention(dim_time, dim_freq, num_species, model_type='feature_level_attention', conv_dim=conv_dim, rnn_dim=rnn_dim, hidden_units=hidden_units, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "\n",
    "loss = CategoricalCrossentropy()\n",
    "# model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=learning_rate_fn), loss=loss)\n",
    "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=learning_rate), loss=loss, metrics=['accuracy'])\n",
    "\n",
    "# class weight\n",
    "weights = compute_class_weight(class_weight='balanced', classes=np.unique(label_train), y=label_train)\n",
    "\n",
    "class_weights = dict()\n",
    "for ii in range(num_species):\n",
    "    class_weights[ii] = weights[ii]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 2.6217 - accuracy: 0.2190\n",
      "Epoch 00001: val_loss improved from inf to 2.14424, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_01_valloss_2.1442_valacc_0.3759.hdf5\n",
      "7145/7145 [==============================] - 232s 32ms/step - loss: 2.6216 - accuracy: 0.2190 - val_loss: 2.1442 - val_accuracy: 0.3759\n",
      "Epoch 2/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 1.9739 - accuracy: 0.3637\n",
      "Epoch 00002: val_loss improved from 2.14424 to 1.77832, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_02_valloss_1.7783_valacc_0.4427.hdf5\n",
      "7145/7145 [==============================] - 232s 32ms/step - loss: 1.9738 - accuracy: 0.3637 - val_loss: 1.7783 - val_accuracy: 0.4427\n",
      "Epoch 3/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 1.6197 - accuracy: 0.4492\n",
      "Epoch 00003: val_loss improved from 1.77832 to 1.56632, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_03_valloss_1.5663_valacc_0.5146.hdf5\n",
      "7145/7145 [==============================] - 232s 32ms/step - loss: 1.6197 - accuracy: 0.4492 - val_loss: 1.5663 - val_accuracy: 0.5146\n",
      "Epoch 4/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 1.3578 - accuracy: 0.5273\n",
      "Epoch 00004: val_loss improved from 1.56632 to 1.34471, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_04_valloss_1.3447_valacc_0.5636.hdf5\n",
      "7145/7145 [==============================] - 232s 32ms/step - loss: 1.3578 - accuracy: 0.5273 - val_loss: 1.3447 - val_accuracy: 0.5636\n",
      "Epoch 5/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 1.1492 - accuracy: 0.5813\n",
      "Epoch 00005: val_loss improved from 1.34471 to 1.09452, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_05_valloss_1.0945_valacc_0.6488.hdf5\n",
      "7145/7145 [==============================] - 233s 33ms/step - loss: 1.1492 - accuracy: 0.5813 - val_loss: 1.0945 - val_accuracy: 0.6488\n",
      "Epoch 6/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 1.0363 - accuracy: 0.6097\n",
      "Epoch 00006: val_loss did not improve from 1.09452\n",
      "7145/7145 [==============================] - 233s 33ms/step - loss: 1.0362 - accuracy: 0.6097 - val_loss: 1.1718 - val_accuracy: 0.6348\n",
      "Epoch 7/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.9343 - accuracy: 0.6376\n",
      "Epoch 00007: val_loss improved from 1.09452 to 0.98839, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_07_valloss_0.9884_valacc_0.6743.hdf5\n",
      "7145/7145 [==============================] - 233s 33ms/step - loss: 0.9343 - accuracy: 0.6376 - val_loss: 0.9884 - val_accuracy: 0.6743\n",
      "Epoch 8/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.8591 - accuracy: 0.6558\n",
      "Epoch 00008: val_loss did not improve from 0.98839\n",
      "7145/7145 [==============================] - 233s 33ms/step - loss: 0.8592 - accuracy: 0.6558 - val_loss: 1.1291 - val_accuracy: 0.6391\n",
      "Epoch 9/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.7933 - accuracy: 0.6728\n",
      "Epoch 00009: val_loss improved from 0.98839 to 0.91626, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_09_valloss_0.9163_valacc_0.6996.hdf5\n",
      "7145/7145 [==============================] - 234s 33ms/step - loss: 0.7933 - accuracy: 0.6728 - val_loss: 0.9163 - val_accuracy: 0.6996\n",
      "Epoch 10/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.7418 - accuracy: 0.6862\n",
      "Epoch 00010: val_loss improved from 0.91626 to 0.79106, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_10_valloss_0.7911_valacc_0.7348.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.7418 - accuracy: 0.6862 - val_loss: 0.7911 - val_accuracy: 0.7348\n",
      "Epoch 11/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.6880 - accuracy: 0.7024\n",
      "Epoch 00011: val_loss did not improve from 0.79106\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.6880 - accuracy: 0.7024 - val_loss: 0.9121 - val_accuracy: 0.7079\n",
      "Epoch 12/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.6523 - accuracy: 0.7107\n",
      "Epoch 00012: val_loss did not improve from 0.79106\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.6523 - accuracy: 0.7107 - val_loss: 0.9492 - val_accuracy: 0.6961\n",
      "Epoch 13/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.6212 - accuracy: 0.7233\n",
      "Epoch 00013: val_loss did not improve from 0.79106\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.6212 - accuracy: 0.7233 - val_loss: 0.8668 - val_accuracy: 0.7166\n",
      "Epoch 14/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.5764 - accuracy: 0.7354\n",
      "Epoch 00014: val_loss did not improve from 0.79106\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.5764 - accuracy: 0.7354 - val_loss: 1.2275 - val_accuracy: 0.6594\n",
      "Epoch 15/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.5746 - accuracy: 0.7379\n",
      "Epoch 00015: val_loss improved from 0.79106 to 0.71079, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_15_valloss_0.7108_valacc_0.7664.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.5746 - accuracy: 0.7379 - val_loss: 0.7108 - val_accuracy: 0.7664\n",
      "Epoch 16/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.5396 - accuracy: 0.7480\n",
      "Epoch 00016: val_loss did not improve from 0.71079\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.5396 - accuracy: 0.7480 - val_loss: 0.8215 - val_accuracy: 0.7350\n",
      "Epoch 17/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.5278 - accuracy: 0.7532\n",
      "Epoch 00017: val_loss did not improve from 0.71079\n",
      "7145/7145 [==============================] - 232s 32ms/step - loss: 0.5278 - accuracy: 0.7532 - val_loss: 0.8495 - val_accuracy: 0.7132\n",
      "Epoch 18/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.4875 - accuracy: 0.7635\n",
      "Epoch 00018: val_loss improved from 0.71079 to 0.65994, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_18_valloss_0.6599_valacc_0.7744.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.4875 - accuracy: 0.7635 - val_loss: 0.6599 - val_accuracy: 0.7744\n",
      "Epoch 19/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.4759 - accuracy: 0.7685\n",
      "Epoch 00019: val_loss did not improve from 0.65994\n",
      "7145/7145 [==============================] - 232s 33ms/step - loss: 0.4759 - accuracy: 0.7685 - val_loss: 0.6904 - val_accuracy: 0.7602\n",
      "Epoch 20/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.4586 - accuracy: 0.7752\n",
      "Epoch 00020: val_loss improved from 0.65994 to 0.62065, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_20_valloss_0.6206_valacc_0.7874.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.4586 - accuracy: 0.7752 - val_loss: 0.6206 - val_accuracy: 0.7874\n",
      "Epoch 21/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.4450 - accuracy: 0.7790\n",
      "Epoch 00021: val_loss improved from 0.62065 to 0.57589, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_21_valloss_0.5759_valacc_0.8007.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.4450 - accuracy: 0.7790 - val_loss: 0.5759 - val_accuracy: 0.8007\n",
      "Epoch 22/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.4114 - accuracy: 0.7904\n",
      "Epoch 00022: val_loss improved from 0.57589 to 0.56705, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_22_valloss_0.5671_valacc_0.8055.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.4114 - accuracy: 0.7904 - val_loss: 0.5671 - val_accuracy: 0.8055\n",
      "Epoch 23/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.4235 - accuracy: 0.7886\n",
      "Epoch 00023: val_loss improved from 0.56705 to 0.55711, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_23_valloss_0.5571_valacc_0.8084.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.4235 - accuracy: 0.7886 - val_loss: 0.5571 - val_accuracy: 0.8084\n",
      "Epoch 24/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.4022 - accuracy: 0.7957\n",
      "Epoch 00024: val_loss did not improve from 0.55711\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.4022 - accuracy: 0.7957 - val_loss: 0.5725 - val_accuracy: 0.8066\n",
      "Epoch 25/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.3747 - accuracy: 0.8032\n",
      "Epoch 00025: val_loss improved from 0.55711 to 0.53765, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_25_valloss_0.5377_valacc_0.8169.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.3747 - accuracy: 0.8032 - val_loss: 0.5377 - val_accuracy: 0.8169\n",
      "Epoch 26/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.3607 - accuracy: 0.8100\n",
      "Epoch 00026: val_loss did not improve from 0.53765\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.3607 - accuracy: 0.8100 - val_loss: 0.5435 - val_accuracy: 0.8118\n",
      "Epoch 27/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.3522 - accuracy: 0.8132\n",
      "Epoch 00027: val_loss improved from 0.53765 to 0.48298, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_27_valloss_0.4830_valacc_0.8353.hdf5\n",
      "7145/7145 [==============================] - 232s 32ms/step - loss: 0.3522 - accuracy: 0.8132 - val_loss: 0.4830 - val_accuracy: 0.8353\n",
      "Epoch 28/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.3386 - accuracy: 0.8183\n",
      "Epoch 00028: val_loss did not improve from 0.48298\n",
      "7145/7145 [==============================] - 233s 33ms/step - loss: 0.3386 - accuracy: 0.8183 - val_loss: 0.5369 - val_accuracy: 0.8182\n",
      "Epoch 29/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.3252 - accuracy: 0.8219\n",
      "Epoch 00029: val_loss did not improve from 0.48298\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.3252 - accuracy: 0.8219 - val_loss: 0.4986 - val_accuracy: 0.8288\n",
      "Epoch 30/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.3312 - accuracy: 0.8225\n",
      "Epoch 00030: val_loss did not improve from 0.48298\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.3312 - accuracy: 0.8225 - val_loss: 0.5029 - val_accuracy: 0.8283\n",
      "Epoch 31/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.3178 - accuracy: 0.8268\n",
      "Epoch 00031: val_loss improved from 0.48298 to 0.47694, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_31_valloss_0.4769_valacc_0.8357.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.3178 - accuracy: 0.8268 - val_loss: 0.4769 - val_accuracy: 0.8357\n",
      "Epoch 32/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.3034 - accuracy: 0.8312\n",
      "Epoch 00032: val_loss improved from 0.47694 to 0.44534, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_32_valloss_0.4453_valacc_0.8449.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.3034 - accuracy: 0.8312 - val_loss: 0.4453 - val_accuracy: 0.8449\n",
      "Epoch 33/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.3051 - accuracy: 0.8332\n",
      "Epoch 00033: val_loss did not improve from 0.44534\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.3051 - accuracy: 0.8332 - val_loss: 0.4725 - val_accuracy: 0.8374\n",
      "Epoch 34/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2872 - accuracy: 0.8384\n",
      "Epoch 00034: val_loss improved from 0.44534 to 0.43827, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_34_valloss_0.4383_valacc_0.8463.hdf5\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2872 - accuracy: 0.8384 - val_loss: 0.4383 - val_accuracy: 0.8463\n",
      "Epoch 35/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2943 - accuracy: 0.8366\n",
      "Epoch 00035: val_loss did not improve from 0.43827\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2943 - accuracy: 0.8366 - val_loss: 0.4476 - val_accuracy: 0.8474\n",
      "Epoch 36/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2682 - accuracy: 0.8463\n",
      "Epoch 00036: val_loss did not improve from 0.43827\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2682 - accuracy: 0.8463 - val_loss: 0.4860 - val_accuracy: 0.8333\n",
      "Epoch 37/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2728 - accuracy: 0.8453\n",
      "Epoch 00037: val_loss did not improve from 0.43827\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2728 - accuracy: 0.8453 - val_loss: 0.4548 - val_accuracy: 0.8463\n",
      "Epoch 38/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2690 - accuracy: 0.8474\n",
      "Epoch 00038: val_loss did not improve from 0.43827\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2690 - accuracy: 0.8474 - val_loss: 0.4517 - val_accuracy: 0.8453\n",
      "Epoch 39/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2528 - accuracy: 0.8528\n",
      "Epoch 00039: val_loss did not improve from 0.43827\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2528 - accuracy: 0.8528 - val_loss: 0.5098 - val_accuracy: 0.8326\n",
      "Epoch 40/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2532 - accuracy: 0.8540\n",
      "Epoch 00040: val_loss improved from 0.43827 to 0.42370, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_40_valloss_0.4237_valacc_0.8556.hdf5\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2532 - accuracy: 0.8540 - val_loss: 0.4237 - val_accuracy: 0.8556\n",
      "Epoch 41/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2457 - accuracy: 0.8571\n",
      "Epoch 00041: val_loss did not improve from 0.42370\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2457 - accuracy: 0.8571 - val_loss: 0.7127 - val_accuracy: 0.7727\n",
      "Epoch 42/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2393 - accuracy: 0.8590\n",
      "Epoch 00042: val_loss improved from 0.42370 to 0.40823, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_42_valloss_0.4082_valacc_0.8582.hdf5\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2393 - accuracy: 0.8590 - val_loss: 0.4082 - val_accuracy: 0.8582\n",
      "Epoch 43/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2534 - accuracy: 0.8586\n",
      "Epoch 00043: val_loss improved from 0.40823 to 0.39849, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_43_valloss_0.3985_valacc_0.8626.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.2534 - accuracy: 0.8586 - val_loss: 0.3985 - val_accuracy: 0.8626\n",
      "Epoch 44/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2352 - accuracy: 0.8625\n",
      "Epoch 00044: val_loss did not improve from 0.39849\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2352 - accuracy: 0.8625 - val_loss: 0.4175 - val_accuracy: 0.8569\n",
      "Epoch 45/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2386 - accuracy: 0.8635\n",
      "Epoch 00045: val_loss improved from 0.39849 to 0.39213, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_45_valloss_0.3921_valacc_0.8650.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.2386 - accuracy: 0.8635 - val_loss: 0.3921 - val_accuracy: 0.8650\n",
      "Epoch 46/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2134 - accuracy: 0.8692\n",
      "Epoch 00046: val_loss improved from 0.39213 to 0.38090, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_46_valloss_0.3809_valacc_0.8674.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.2134 - accuracy: 0.8692 - val_loss: 0.3809 - val_accuracy: 0.8674\n",
      "Epoch 47/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2052 - accuracy: 0.8737\n",
      "Epoch 00047: val_loss improved from 0.38090 to 0.36417, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_47_valloss_0.3642_valacc_0.8733.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.2052 - accuracy: 0.8737 - val_loss: 0.3642 - val_accuracy: 0.8733\n",
      "Epoch 48/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2077 - accuracy: 0.8728\n",
      "Epoch 00048: val_loss did not improve from 0.36417\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2077 - accuracy: 0.8728 - val_loss: 0.4690 - val_accuracy: 0.8408\n",
      "Epoch 49/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2036 - accuracy: 0.8755\n",
      "Epoch 00049: val_loss did not improve from 0.36417\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2036 - accuracy: 0.8755 - val_loss: 0.3813 - val_accuracy: 0.8680\n",
      "Epoch 50/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.2065 - accuracy: 0.8755\n",
      "Epoch 00050: val_loss did not improve from 0.36417\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.2065 - accuracy: 0.8755 - val_loss: 0.3706 - val_accuracy: 0.8725\n",
      "Epoch 51/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1935 - accuracy: 0.8808\n",
      "Epoch 00051: val_loss did not improve from 0.36417\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1935 - accuracy: 0.8808 - val_loss: 0.4068 - val_accuracy: 0.8607\n",
      "Epoch 52/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1861 - accuracy: 0.8842\n",
      "Epoch 00052: val_loss improved from 0.36417 to 0.36188, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_52_valloss_0.3619_valacc_0.8762.hdf5\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1861 - accuracy: 0.8842 - val_loss: 0.3619 - val_accuracy: 0.8762\n",
      "Epoch 53/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1978 - accuracy: 0.8803\n",
      "Epoch 00053: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1978 - accuracy: 0.8803 - val_loss: 0.3919 - val_accuracy: 0.8678\n",
      "Epoch 54/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1965 - accuracy: 0.8808\n",
      "Epoch 00054: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1965 - accuracy: 0.8808 - val_loss: 0.3738 - val_accuracy: 0.8708\n",
      "Epoch 55/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1766 - accuracy: 0.8868\n",
      "Epoch 00055: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1766 - accuracy: 0.8868 - val_loss: 0.3742 - val_accuracy: 0.8742\n",
      "Epoch 56/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.1849 - accuracy: 0.8855\n",
      "Epoch 00056: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.1849 - accuracy: 0.8855 - val_loss: 0.3750 - val_accuracy: 0.8732\n",
      "Epoch 57/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1768 - accuracy: 0.8892\n",
      "Epoch 00057: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1768 - accuracy: 0.8892 - val_loss: 0.3687 - val_accuracy: 0.8769\n",
      "Epoch 58/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.1692 - accuracy: 0.8905\n",
      "Epoch 00058: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1692 - accuracy: 0.8905 - val_loss: 0.3639 - val_accuracy: 0.8788\n",
      "Epoch 59/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1707 - accuracy: 0.8917\n",
      "Epoch 00059: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1707 - accuracy: 0.8917 - val_loss: 0.3989 - val_accuracy: 0.8699\n",
      "Epoch 60/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1644 - accuracy: 0.8947\n",
      "Epoch 00060: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1644 - accuracy: 0.8947 - val_loss: 0.3715 - val_accuracy: 0.8736\n",
      "Epoch 61/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1701 - accuracy: 0.8941\n",
      "Epoch 00061: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1701 - accuracy: 0.8941 - val_loss: 0.3689 - val_accuracy: 0.8773\n",
      "Epoch 62/200\n",
      "7144/7145 [============================>.] - ETA: 0s - loss: 0.1599 - accuracy: 0.8973\n",
      "Epoch 00062: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1599 - accuracy: 0.8973 - val_loss: 0.3830 - val_accuracy: 0.8770\n",
      "Epoch 63/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1699 - accuracy: 0.8969\n",
      "Epoch 00063: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1699 - accuracy: 0.8969 - val_loss: 0.3804 - val_accuracy: 0.8723\n",
      "Epoch 64/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1612 - accuracy: 0.8975\n",
      "Epoch 00064: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1612 - accuracy: 0.8975 - val_loss: 0.3945 - val_accuracy: 0.8719\n",
      "Epoch 65/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1516 - accuracy: 0.9002\n",
      "Epoch 00065: val_loss did not improve from 0.36188\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1516 - accuracy: 0.9002 - val_loss: 0.3876 - val_accuracy: 0.8726\n",
      "Epoch 66/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1597 - accuracy: 0.8994\n",
      "Epoch 00066: val_loss improved from 0.36188 to 0.35607, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_66_valloss_0.3561_valacc_0.8794.hdf5\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1597 - accuracy: 0.8994 - val_loss: 0.3561 - val_accuracy: 0.8794\n",
      "Epoch 67/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1450 - accuracy: 0.9039\n",
      "Epoch 00067: val_loss improved from 0.35607 to 0.34704, saving model to /home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_67_valloss_0.3470_valacc_0.8832.hdf5\n",
      "7145/7145 [==============================] - 231s 32ms/step - loss: 0.1450 - accuracy: 0.9039 - val_loss: 0.3470 - val_accuracy: 0.8832\n",
      "Epoch 68/200\n",
      "7145/7145 [==============================] - ETA: 0s - loss: 0.1530 - accuracy: 0.9021\n",
      "Epoch 00068: val_loss did not improve from 0.34704\n",
      "7145/7145 [==============================] - 230s 32ms/step - loss: 0.1530 - accuracy: 0.9021 - val_loss: 0.4954 - val_accuracy: 0.8453\n",
      "Epoch 69/200\n",
      "5747/7145 [=======================>......] - ETA: 40s - loss: 0.1442 - accuracy: 0.9059"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-25-53c72438cf53>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0;31m# history = model.fit(fea_train, to_categorical(label_train), class_weight=class_weights, validation_split=0.3, batch_size=batch_size, epochs=num_epoch, callbacks=[EarlyStopping(patience=num_patience), ModelCheckpoint(filepath=os.path.join(fit_result_path, '{epoch:02d}-{val_loss:.4f}.hdf5'), verbose=1, monitor=\"val_loss\", save_best_only=True)])\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;31m# history = model.fit(train_generator, validation_data=validate_generator, class_weight=class_weights, epochs=num_epoch, callbacks=[EarlyStopping(patience=num_patience, monitor='val_loss', mode='min', verbose=1), TensorBoard(log_dir=fit_result_path1), ModelCheckpoint(filepath=os.path.join(fit_result_path1, '{epoch:02d}-{val_loss:.4f}.hdf5'), verbose=1, monitor=\"val_loss\", save_best_only=True)])\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 4\u001b[0;31m \u001b[0mhistory\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtrain_generator\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalidation_data\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidate_generator\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mclass_weight\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mclass_weights\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepochs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mnum_epoch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcallbacks\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mEarlyStopping\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpatience\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mnum_patience\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmonitor\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'val_loss'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmode\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'min'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mTensorBoard\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlog_dir\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mfit_result_path1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mModelCheckpoint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfilepath\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mos\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mjoin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfit_result_path1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'epoch_{epoch:02d}_valloss_{val_loss:.4f}_valacc_{val_accuracy:.4f}.hdf5'\u001b[0m \u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmonitor\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m\"val_loss\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msave_best_only\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py\u001b[0m in \u001b[0;36m_method_wrapper\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m     64\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_method_wrapper\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     65\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_in_multi_worker_mode\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m  \u001b[0;31m# pylint: disable=protected-access\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 66\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mmethod\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     67\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     68\u001b[0m     \u001b[0;31m# Running inside `run_distribute_coordinator` already.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)\u001b[0m\n\u001b[1;32m    853\u001b[0m                 \u001b[0mcontext\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0masync_wait\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    854\u001b[0m               \u001b[0mlogs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtmp_logs\u001b[0m  \u001b[0;31m# No error, now safe to assign to logs.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 855\u001b[0;31m               \u001b[0mcallbacks\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mon_train_batch_end\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mstep\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlogs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    856\u001b[0m         \u001b[0mepoch_logs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcopy\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcopy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlogs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    857\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/keras/callbacks.py\u001b[0m in \u001b[0;36mon_train_batch_end\u001b[0;34m(self, batch, logs)\u001b[0m\n\u001b[1;32m    387\u001b[0m     \"\"\"\n\u001b[1;32m    388\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_should_call_train_batch_hooks\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 389\u001b[0;31m       \u001b[0mlogs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_process_logs\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlogs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    390\u001b[0m       \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call_batch_hook\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mModeKeys\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTRAIN\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'end'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlogs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mlogs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    391\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/keras/callbacks.py\u001b[0m in \u001b[0;36m_process_logs\u001b[0;34m(self, logs)\u001b[0m\n\u001b[1;32m    263\u001b[0m     \u001b[0;34m\"\"\"Turns tensors into numpy arrays or Python scalars.\"\"\"\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    264\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mlogs\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 265\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mtf_utils\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mto_numpy_or_python_type\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlogs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    266\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0;34m{\u001b[0m\u001b[0;34m}\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    267\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/keras/utils/tf_utils.py\u001b[0m in \u001b[0;36mto_numpy_or_python_type\u001b[0;34m(tensors)\u001b[0m\n\u001b[1;32m    521\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mt\u001b[0m  \u001b[0;31m# Don't turn ragged or sparse tensors to NumPy.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    522\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 523\u001b[0;31m   \u001b[0;32mreturn\u001b[0m \u001b[0mnest\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmap_structure\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0m_to_single_numpy_or_python_type\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtensors\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    524\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/util/nest.py\u001b[0m in \u001b[0;36mmap_structure\u001b[0;34m(func, *structure, **kwargs)\u001b[0m\n\u001b[1;32m    615\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    616\u001b[0m   return pack_sequence_as(\n\u001b[0;32m--> 617\u001b[0;31m       \u001b[0mstructure\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mx\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mentries\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    618\u001b[0m       expand_composites=expand_composites)\n\u001b[1;32m    619\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/util/nest.py\u001b[0m in \u001b[0;36m<listcomp>\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m    615\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    616\u001b[0m   return pack_sequence_as(\n\u001b[0;32m--> 617\u001b[0;31m       \u001b[0mstructure\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mx\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mentries\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    618\u001b[0m       expand_composites=expand_composites)\n\u001b[1;32m    619\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/keras/utils/tf_utils.py\u001b[0m in \u001b[0;36m_to_single_numpy_or_python_type\u001b[0;34m(t)\u001b[0m\n\u001b[1;32m    517\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_to_single_numpy_or_python_type\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mt\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    518\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mt\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTensor\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 519\u001b[0;31m       \u001b[0mx\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnumpy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    520\u001b[0m       \u001b[0;32mreturn\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mitem\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mndim\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;36m0\u001b[0m \u001b[0;32melse\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    521\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mt\u001b[0m  \u001b[0;31m# Don't turn ragged or sparse tensors to NumPy.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36mnumpy\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    959\u001b[0m     \"\"\"\n\u001b[1;32m    960\u001b[0m     \u001b[0;31m# TODO(slebedev): Consider avoiding a copy for non-CPU or remote tensors.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 961\u001b[0;31m     \u001b[0mmaybe_arr\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_numpy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m  \u001b[0;31m# pylint: disable=protected-access\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    962\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mmaybe_arr\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcopy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmaybe_arr\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mndarray\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32melse\u001b[0m \u001b[0mmaybe_arr\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    963\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/whistle_classifier/lib/python3.8/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36m_numpy\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    925\u001b[0m     \u001b[0;31m# pylint: disable=protected-access\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    926\u001b[0m     \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 927\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_numpy_internal\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    928\u001b[0m     \u001b[0;32mexcept\u001b[0m \u001b[0mcore\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_NotOkStatusException\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    929\u001b[0m       \u001b[0msix\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mraise_from\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcore\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_status_to_exception\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcode\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmessage\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "# With classes\n",
    "# history = model.fit(fea_train, to_categorical(label_train), class_weight=class_weights, validation_split=0.3, batch_size=batch_size, epochs=num_epoch, callbacks=[EarlyStopping(patience=num_patience), ModelCheckpoint(filepath=os.path.join(fit_result_path, '{epoch:02d}-{val_loss:.4f}.hdf5'), verbose=1, monitor=\"val_loss\", save_best_only=True)])\n",
    "# history = model.fit(train_generator, validation_data=validate_generator, class_weight=class_weights, epochs=num_epoch, callbacks=[EarlyStopping(patience=num_patience, monitor='val_loss', mode='min', verbose=1), TensorBoard(log_dir=fit_result_path1), ModelCheckpoint(filepath=os.path.join(fit_result_path1, '{epoch:02d}-{val_loss:.4f}.hdf5'), verbose=1, monitor=\"val_loss\", save_best_only=True)])\n",
    "history = model.fit(train_generator, validation_data=validate_generator, class_weight=class_weights, epochs=num_epoch, callbacks=[EarlyStopping(patience=num_patience, monitor='val_loss', mode='min', verbose=1), TensorBoard(log_dir=fit_result_path1), ModelCheckpoint(filepath=os.path.join(fit_result_path1, 'epoch_{epoch:02d}_valloss_{val_loss:.4f}_valacc_{val_accuracy:.4f}.hdf5' ), verbose=1, monitor=\"val_loss\", save_best_only=True)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best model:0.347\n",
      "/home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210211_120511/epoch_67_valloss_0.3470_valacc_0.8832.hdf5\n"
     ]
    }
   ],
   "source": [
    "# fit_result_path1 = '/home/ys587/__Data/__whistle/__whislte_30_species/__fit_result_species/20210210_224527'\n",
    "the_best_model, _ = find_best_model(fit_result_path1, purge=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input (InputLayer)              [(None, 100, 128, 1) 0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d (Conv2D)                 (None, 100, 128, 16) 160         input[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization (BatchNorma (None, 100, 128, 16) 64          conv2d[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "activation (Activation)         (None, 100, 128, 16) 0           batch_normalization[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_1 (Conv2D)               (None, 100, 128, 16) 2320        activation[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_1 (BatchNor (None, 100, 128, 16) 64          conv2d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_1 (Activation)       (None, 100, 128, 16) 0           batch_normalization_1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D)    (None, 50, 64, 16)   0           activation_1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "dropout (Dropout)               (None, 50, 64, 16)   0           max_pooling2d[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_2 (Conv2D)               (None, 50, 64, 32)   4640        dropout[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_2 (BatchNor (None, 50, 64, 32)   128         conv2d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_2 (Activation)       (None, 50, 64, 32)   0           batch_normalization_2[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_3 (Conv2D)               (None, 50, 64, 32)   9248        activation_2[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_3 (BatchNor (None, 50, 64, 32)   128         conv2d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_3 (Activation)       (None, 50, 64, 32)   0           batch_normalization_3[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2D)  (None, 25, 32, 32)   0           activation_3[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "dropout_1 (Dropout)             (None, 25, 32, 32)   0           max_pooling2d_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_4 (Conv2D)               (None, 25, 32, 64)   18496       dropout_1[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_4 (BatchNor (None, 25, 32, 64)   256         conv2d_4[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_4 (Activation)       (None, 25, 32, 64)   0           batch_normalization_4[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_5 (Conv2D)               (None, 25, 32, 64)   36928       activation_4[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_5 (BatchNor (None, 25, 32, 64)   256         conv2d_5[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_5 (Activation)       (None, 25, 32, 64)   0           batch_normalization_5[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2D)  (None, 13, 16, 64)   0           activation_5[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "dropout_2 (Dropout)             (None, 13, 16, 64)   0           max_pooling2d_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_6 (Conv2D)               (None, 13, 16, 128)  73856       dropout_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_6 (BatchNor (None, 13, 16, 128)  512         conv2d_6[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_6 (Activation)       (None, 13, 16, 128)  0           batch_normalization_6[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_7 (Conv2D)               (None, 13, 16, 128)  147584      activation_6[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_7 (BatchNor (None, 13, 16, 128)  512         conv2d_7[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_7 (Activation)       (None, 13, 16, 128)  0           batch_normalization_7[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2D)  (None, 7, 8, 128)    0           activation_7[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "dropout_3 (Dropout)             (None, 7, 8, 128)    0           max_pooling2d_3[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_8 (Conv2D)               (None, 7, 8, 256)    295168      dropout_3[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_8 (BatchNor (None, 7, 8, 256)    1024        conv2d_8[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_8 (Activation)       (None, 7, 8, 256)    0           batch_normalization_8[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_9 (Conv2D)               (None, 7, 8, 256)    590080      activation_8[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_9 (BatchNor (None, 7, 8, 256)    1024        conv2d_9[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "activation_9 (Activation)       (None, 7, 8, 256)    0           batch_normalization_9[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2D)  (None, 7, 4, 256)    0           activation_9[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "dropout_4 (Dropout)             (None, 7, 4, 256)    0           max_pooling2d_4[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_10 (Conv2D)              (None, 7, 4, 512)    1180160     dropout_4[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_10 (BatchNo (None, 7, 4, 512)    2048        conv2d_10[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "activation_10 (Activation)      (None, 7, 4, 512)    0           batch_normalization_10[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_11 (Conv2D)              (None, 7, 4, 512)    2359808     activation_10[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_11 (BatchNo (None, 7, 4, 512)    2048        conv2d_11[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "activation_11 (Activation)      (None, 7, 4, 512)    0           batch_normalization_11[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_5 (MaxPooling2D)  (None, 7, 2, 512)    0           activation_11[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dropout_5 (Dropout)             (None, 7, 2, 512)    0           max_pooling2d_5[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "reshape (Reshape)               (None, 7, 1024)      0           dropout_5[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense (Dense)                   (None, 7, 512)       524800      reshape[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_12 (BatchNo (None, 7, 512)       2048        dense[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "activation_12 (Activation)      (None, 7, 512)       0           batch_normalization_12[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_6 (Dropout)             (None, 7, 512)       0           activation_12[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 7, 512)       262656      dropout_6[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_13 (BatchNo (None, 7, 512)       2048        dense_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "activation_13 (Activation)      (None, 7, 512)       0           batch_normalization_13[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_7 (Dropout)             (None, 7, 512)       0           activation_13[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 7, 512)       262656      dropout_7[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_14 (BatchNo (None, 7, 512)       2048        dense_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "activation_14 (Activation)      (None, 7, 512)       0           batch_normalization_14[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_8 (Dropout)             (None, 7, 512)       0           activation_14[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_3 (Dense)                 (None, 7, 512)       262656      dropout_8[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_4 (Dense)                 (None, 7, 512)       262656      dropout_8[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "lambda (Lambda)                 (None, 512)          0           dense_3[0][0]                    \n",
      "                                                                 dense_4[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_15 (BatchNo (None, 512)          2048        lambda[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "activation_15 (Activation)      (None, 512)          0           batch_normalization_15[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_9 (Dropout)             (None, 512)          0           activation_15[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_5 (Dense)                 (None, 21)           10773       dropout_9[0][0]                  \n",
      "==================================================================================================\n",
      "Total params: 6,320,901\n",
      "Trainable params: 6,312,773\n",
      "Non-trainable params: 8,128\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = load_model(the_best_model)\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_pred = model.predict(fea_test)\n",
    "# label_pred = model.predict(test_generator)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.set_printoptions(linewidth=200, precision=2, suppress=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix:\n",
      "['BD', 'MH', 'CD', 'STR', 'SPT', 'SPIN', 'PLT', 'RD', 'RT', 'WSD', 'FKW', 'BEL', 'KW', 'WBD', 'DUSK', 'FRA', 'PKW', 'LPLT', 'CLY', 'SPE', 'ASP']\n",
      "\n",
      "[[  39    2   12    9   23   12    3    0    3    9    2    0    0    0    0    4    0    2    0    1    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [ 621  455 1764  386  308  121  160   12   24   38   13    1    1    6    0   23    1   15    0    4   11]\n",
      " [ 298   29  194  202  217   70   66    1   21   12    5    0    0    5    0   13    0    0    1    3    3]\n",
      " [  67   75  500   15   99   53   15    0    2    5    1    0    0    1    0    5    0    0    6    1    0]\n",
      " [ 155    3   26   30  111  148    3    0    0    1    0    0    0    0    1    2    0    0    1    0   10]\n",
      " [   0    0    0    0    0    0   24    0    0    0    4    0    0    0    0    0    0    3    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [  38    8    2    4    1    0    4    2   11    3    2    0    0    0    0    1    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]\n",
      " [   0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0    0]]\n",
      "\n",
      "[[0.32 0.02 0.1  0.07 0.19 0.1  0.02 0.   0.02 0.07 0.02 0.   0.   0.   0.   0.03 0.   0.02 0.   0.01 0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.16 0.11 0.45 0.1  0.08 0.03 0.04 0.   0.01 0.01 0.   0.   0.   0.   0.   0.01 0.   0.   0.   0.   0.  ]\n",
      " [0.26 0.03 0.17 0.18 0.19 0.06 0.06 0.   0.02 0.01 0.   0.   0.   0.   0.   0.01 0.   0.   0.   0.   0.  ]\n",
      " [0.08 0.09 0.59 0.02 0.12 0.06 0.02 0.   0.   0.01 0.   0.   0.   0.   0.   0.01 0.   0.   0.01 0.   0.  ]\n",
      " [0.32 0.01 0.05 0.06 0.23 0.3  0.01 0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.02]\n",
      " [0.   0.   0.   0.   0.   0.   0.77 0.   0.   0.   0.13 0.   0.   0.   0.   0.   0.   0.1  0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.5  0.11 0.03 0.05 0.01 0.   0.05 0.03 0.14 0.04 0.03 0.   0.   0.   0.   0.01 0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]\n",
      " [0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.   0.  ]]\n"
     ]
    }
   ],
   "source": [
    "print(\"Confusion matrix:\")\n",
    "cm = confusion_matrix(label_test, np.argmax(label_pred, axis=1), labels=species_id)\n",
    "\n",
    "print(species_list)\n",
    "print('')\n",
    "print(cm)\n",
    "print('')\n",
    "\n",
    "cm2 = cm*1.0\n",
    "for ii in range(cm.shape[0]):\n",
    "    cm_row = cm[ii, :]*1.0\n",
    "\n",
    "    cm_row_sum = cm_row.sum()\n",
    "    if cm_row_sum != 0:\n",
    "        cm2[ii, :] = cm_row / cm_row_sum\n",
    "    else:\n",
    "        cm2[ii, :] = np.zeros(cm.shape[1])\n",
    "\n",
    "print(cm2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import plot_confusion_matrix\n",
    "from sklearn.metrics._plot.confusion_matrix import ConfusionMatrixDisplay\n",
    "disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=species_list)\n",
    "disp2 = ConfusionMatrixDisplay(confusion_matrix=cm2, display_labels=species_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7f80d4e02d00>"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x1080 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots(figsize=[15, 15])\n",
    "disp.plot(include_values=True,\n",
    "                     cmap='viridis', ax=ax, xticks_rotation='horizontal',\n",
    "                     values_format=None, colorbar=True)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7f80d47a7ee0>"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x1080 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots(figsize=[15, 15])\n",
    "disp2.plot(include_values=True,\n",
    "                     cmap='viridis', ax=ax, xticks_rotation='horizontal',\n",
    "                     values_format='.2f', colorbar=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# part of training data\n",
    "# Testing data\n",
    "fea_temp = np.load(os.path.join(feature_path, 'oswald_part1_orig.npz'))\n",
    "fea_test2 = fea_temp['feas_orig']\n",
    "label_test_list2 = fea_temp['labels_orig']\n",
    "\n",
    "fea_test2 = fea_test2[:,:100,:]\n",
    "label_test2 = np.zeros(len(label_test_list2))\n",
    "for ii in range(len(label_test_list2)):\n",
    "    label_test2[ii] = species_dict[label_test_list2[ii]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "the_best_model, _ = find_best_model(fit_result_path1, purge=False)\n",
    "del model\n",
    "model = model_cnn14_attention_multi(dim_time, dim_freq, num_species, model_type='feature_level_attention', conv_dim=conv_dim, pool_size=pool_size, pool_stride=pool_stride, hidden_units=hidden_units, l2_regu=l2_regu, drop_rate=drop_rate)\n",
    "model = load_model(the_best_model)\n",
    "label_pred2 = model.predict(fea_test2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(\"Confusion matrix:\")\n",
    "cm = confusion_matrix(label_test2, np.argmax(label_pred2, axis=1))\n",
    "    \n",
    "print(species_list)\n",
    "print('')\n",
    "print(cm)\n",
    "print('')\n",
    "\n",
    "cm2 = cm*1.0\n",
    "for ii in range(cm.shape[0]):\n",
    "    cm_row = cm[ii, :]*1.0\n",
    "\n",
    "    cm_row_sum = cm_row.sum()\n",
    "    if cm_row_sum != 0:\n",
    "        cm2[ii, :] = cm_row / cm_row_sum\n",
    "    else:\n",
    "        cm2[ii, :] = np.zeros(cm.shape[1])\n",
    "\n",
    "print(cm2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "from IPython.utils.io import Tee\n",
    "from contextlib import closing\n",
    "\n",
    "import sys\n",
    "import numpy\n",
    "numpy.set_printoptions(threshold=sys.maxsize)\n",
    "\n",
    "with closing(Tee(os.path.join(fit_result_path1, 'net_architecture.txt'), \"w\", channel=\"stdout\")) as outputstream:\n",
    "    print(\"Detailed classification report:\")\n",
    "    print()\n",
    "    \n",
    "    # print('learning_rate = {:.2e}'.format(learning_rate))\n",
    "    print('l2_regu = {:.2e}'.format(l2_regu))\n",
    "    print('drop_rate = {:.3f}'.format(drop_rate))\n",
    "    print('conv_dim = {:d}'.format(conv_dim))\n",
    "    print('rnn_dim = {:d}'.format(rnn_dim))\n",
    "    print('pool_size = {:d}'.format(pool_size))\n",
    "    print('pool_stride = {:d}'.format(pool_stride))\n",
    "    print('hidden_units = {:d}'.format(hidden_units))\n",
    "    print('fcn_dim = {:d}'.format(fcn_dim))\n",
    "    \n",
    "    print()\n",
    "    print(the_best_model)\n",
    "    print()\n",
    "    # print(classification_report(label_test, np.argmax(label_pred, axis=1), target_names=species_id))\n",
    "    print(classification_report(label_test, np.argmax(label_pred, axis=1) ))\n",
    "    print()\n",
    "\n",
    "    print(\"Confusion matrix:\")\n",
    "    cm = confusion_matrix(label_test, np.argmax(label_pred, axis=1))\n",
    "    print(cm)\n",
    "\n",
    "    cm2 = cm*1.0\n",
    "    for ii in range(cm.shape[0]):\n",
    "        cm_row = cm[ii, :]*1.0\n",
    "        \n",
    "        cm_row_sum = cm_row.sum()\n",
    "        if cm_row != 0:\n",
    "            cm2[ii, :] = cm_row / cm_row_sum\n",
    "        else:\n",
    "            cm2[ii, :] = np.zeros(cm.shape[1])\n",
    "    print()\n",
    "    print(np.around(cm2, 3))\n",
    "\n",
    "    model.summary()\n",
    "    "
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "              precision    recall  f1-score   support\n",
    "\n",
    "         0.0       0.95      0.88      0.91      1639\n",
    "         1.0       0.00      0.00      0.00         0\n",
    "         2.0       0.97      0.89      0.93      3964\n",
    "         3.0       0.76      0.95      0.84      1512\n",
    "         4.0       0.94      0.94      0.94      1327\n",
    "         5.0       0.87      0.86      0.87       896\n",
    "         6.0       0.99      0.99      0.99       323\n",
    "         8.0       0.96      0.99      0.97       516\n",
    "         9.0       0.00      0.00      0.00         0\n",
    "        10.0       0.99      0.99      0.99      2007\n",
    "        18.0       0.00      0.00      0.00         0\n",
    "        20.0       0.00      0.00      0.00         0\n",
    "\n",
    "    accuracy                           0.93     12184\n",
    "   macro avg       0.62      0.63      0.62     12184\n",
    "weighted avg       0.93      0.93      0.93     12184\n",
    "\n",
    "\n",
    "Confusion matrix:\n",
    "[[1438    4   12   91   23   49    2    6    1   13    0    0]\n",
    " [   0    0    0    0    0    0    0    0    0    0    0    0]\n",
    " [  23   21 3544  269   45   45    0    8    5    3    0    1]\n",
    " [   6    2   52 1436    6    7    0    2    1    0    0    0]\n",
    " [  12    0    7   39 1252   15    0    1    0    0    1    0]\n",
    " [  35    0   28   52    7  774    0    0    0    0    0    0]\n",
    " [   1    0    0    1    0    0  320    0    0    1    0    0]\n",
    " [   0    0    1    1    0    0    1  512    0    1    0    0]\n",
    " [   0    0    0    0    0    0    0    0    0    0    0    0]\n",
    " [   0    5    1    0    0    0    0    6    0 1995    0    0]\n",
    " [   0    0    0    0    0    0    0    0    0    0    0    0]\n",
    " [   0    0    0    0    0    0    0    0    0    0    0    0]]\n",
    "\n",
    "[[0.877 0.002 0.007 0.056 0.014 0.03  0.001 0.004 0.001 0.008 0.    0.   ]\n",
    " [  nan   nan   nan   nan   nan   nan   nan   nan   nan   nan   nan   nan]\n",
    " [0.006 0.005 0.894 0.068 0.011 0.011 0.    0.002 0.001 0.001 0.    0.   ]\n",
    " [0.004 0.001 0.034 0.95  0.004 0.005 0.    0.001 0.001 0.    0.    0.   ]\n",
    " [0.009 0.    0.005 0.029 0.943 0.011 0.    0.001 0.    0.    0.001 0.   ]\n",
    " [0.039 0.    0.031 0.058 0.008 0.864 0.    0.    0.    0.    0.    0.   ]\n",
    " [0.003 0.    0.    0.003 0.    0.    0.991 0.    0.    0.003 0.    0.   ]\n",
    " [0.    0.    0.002 0.002 0.    0.    0.002 0.992 0.    0.002 0.    0.   ]\n",
    " [  nan   nan   nan   nan   nan   nan   nan   nan   nan   nan   nan   nan]\n",
    " [0.    0.002 0.    0.    0.    0.    0.    0.003 0.    0.994 0.    0.   ]\n",
    " [  nan   nan   nan   nan   nan   nan   nan   nan   nan   nan   nan   nan]\n",
    " [  nan   nan   nan   nan   nan   nan   nan   nan   nan   nan   nan   nan]]\n",
    "Model: \"model\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## learning rate scheduler\n",
    "step = tf.Variable(0, trainable=False)\n",
    "boundaries = [10, 20]\n",
    "values = [1.0e-4, 3.3e-5, 1.0e-5]\n",
    "learning_rate_fn = PiecewiseConstantDecay(boundaries, values)\n",
    "\n",
    "# Later, whenever we perform an optimization step, we pass in the step.\n",
    "# learning_rate = learning_rate_fn(step)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### if False:\n",
    "    # fea_files = ['oswald_part1_orig.npz', 'oswald_part1_aug.npz', 'oswald_part2_orig.npz', 'oswald_part2_aug.npz',\n",
    "    #             'gillispie_48kHz_orig.npz', 'gillispie_48kHz_aug.npz', 'gillispie_96kHz_orig.npz', 'gillispie_48kHz_aug.npz',\n",
    "    #             'dclde2011_orig.npz', 'dclde2011_aug.npz', 'watkin_orig.npz', 'watkin_aug.npz']\n",
    "\n",
    "    # Expt 1: test on oswald_part1\n",
    "    fea_train_files = ['oswald_part1_orig.npz', 'oswald_part1_aug.npz',\n",
    "                'gillispie_48kHz_orig.npz', 'gillispie_48kHz_aug.npz', 'gillispie_96kHz_orig.npz', 'gillispie_96kHz_aug.npz',\n",
    "                'dclde2011_orig.npz', 'dclde2011_aug.npz', 'watkin_orig.npz', 'watkin_aug.npz']\n",
    "\n",
    "    fea_test_files = ['oswald_part2_orig.npz']\n",
    "\n",
    "    # Training data\n",
    "    fea_train_list = []\n",
    "    label_train_list = []\n",
    "    for ii in range(len(fea_train_files)):\n",
    "        ff = fea_train_files[ii]\n",
    "        print(ff)\n",
    "        fea_temp = np.load(os.path.join(feature_path, ff))\n",
    "        print(fea_temp.files)\n",
    "\n",
    "        if ii == 0:\n",
    "            fea_train = fea_temp['feas_orig']\n",
    "            label_train = fea_temp['labels_orig']\n",
    "            print(fea_train.shape)\n",
    "            print(label_train.shape)\n",
    "        elif ii % 2 == 0:  # even\n",
    "            fea_train = np.concatenate([fea_train, fea_temp['feas_orig']])\n",
    "            label_train = np.concatenate([label_train, fea_temp['labels_orig']])\n",
    "            print(fea_train.shape)\n",
    "            print(label_train.shape)\n",
    "        else:\n",
    "            fea_train = np.concatenate([fea_train, fea_temp['feas_aug']])\n",
    "            label_train = np.concatenate([label_train, fea_temp['labels_aug']])\n",
    "            print(fea_train.shape)\n",
    "            print(label_train.shape)\n",
    "        print('')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}