my_bengio.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "names = open('data/names.txt','r').read()\n",
    "import matplotlib.pyplot as plt # for making figures\n",
    "%matplotlib inline\n",
    "t= torch.tensor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "SPECIAL = '.'\n",
    "names = names.replace('\\n','.')\n",
    "\n",
    "chars = sorted(list(set(''.join(names))))\n",
    "stoi = {s:i for i,s in enumerate(chars)}\n",
    "itos = {i:s for s,i in stoi.items()}\n",
    "\n",
    "\n",
    "\n",
    "def label_data(context_size, names):\n",
    "    X = []\n",
    "    Y = []\n",
    "    context = [0] * context_size    \n",
    "\n",
    "    \n",
    "    for letter in names + '.':\n",
    "        letter_as_int = stoi[letter]\n",
    "        X.append(context)\n",
    "        Y.append(letter_as_int)\n",
    "        context = context[1:] + [letter_as_int]\n",
    "    \n",
    "\n",
    "    return (X,Y) \n",
    "\n",
    "X,Y = label_data(3, names)\n",
    "\n",
    "def split_data(data):\n",
    "    data_size = len(data)\n",
    "    dev_size = int(data_size*0.8)\n",
    "    tr_size = int(data_size*0.1)\n",
    "    \n",
    "    dev = t(data[:dev_size])\n",
    "    tr = t(data[dev_size:dev_size + tr_size])\n",
    "    tst = t(data[dev_size+tr_size:])\n",
    "    \n",
    "    return (dev,tr,tst)\n",
    "\n",
    "X_dev,X_tr,X_tst = split_data(X)\n",
    "Y_dev,Y_tr,Y_tst = split_data(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "params count: 6097\n"
     ]
    }
   ],
   "source": [
    "# let embeding C E R^m, where m = 2, |V| = 27\n",
    "g = torch.Generator().manual_seed(2147483647) # for reproducibility\n",
    "C = torch.randn((27, 10), generator=g)\n",
    "W1 = torch.randn((30, 100), generator=g)\n",
    "B1 = torch.randn(100, generator=g)\n",
    "W2 = torch.randn((100, 27), generator=g)\n",
    "B2 = torch.randn(27, generator=g)\n",
    "parameters = [C, W1, B1, W2, B2]\n",
    "\n",
    "params = [C,W1,B1,W2,B2]\n",
    "print(\"params count:\",sum(p.nelement() for p in parameters))\n",
    "\n",
    "for p in params:\n",
    "    p.requires_grad = True\n",
    "\n",
    "i = 0\n",
    "lri = []\n",
    "\n",
    "def step(X,Y,lr):\n",
    "    global i\n",
    "    global lri\n",
    "    ix = torch.randint(0, X.shape[0], (32,))\n",
    "\n",
    "    emb = C[X[ix]]\n",
    "    emb_aligned = emb.view(-1,30)\n",
    "    # layer 1 \n",
    "    for param in params:\n",
    "        param.grad = None\n",
    "    \n",
    "    layer_1 = torch.tanh(emb_aligned @ W1 + B1)\n",
    "    layer_2 = layer_1 @ W2 + B2\n",
    "    loss = torch.nn.functional.cross_entropy(layer_2,Y[ix])\n",
    "   \n",
    "    loss.backward()\n",
    "\n",
    "    for param in params:\n",
    "        param.data -= lr * param.grad\n",
    "    \n",
    "    return loss.item()\n",
    "\n",
    "def evaluate(X,Y):\n",
    "    emb = C[X]\n",
    "\n",
    "    layer_1 = torch.tanh(emb.view(-1,30) @ W1 + B1)\n",
    "    layer_2 = layer_1 @ W2 + B2\n",
    "    return torch.nn.functional.cross_entropy(layer_2,Y)\n",
    "\n",
    "\n",
    "\n",
    "# lre = torch.linspace(-4,0, 1000)\n",
    "# lrs = 10 ** lre\n",
    "# for i in range(1000):\n",
    "#     loss = step(X_dev,Y_dev, lrs[i])\n",
    "#     lri.append(loss)\n",
    "# plt.plot(lre,lri)\n",
    "# => \n",
    "lr = 10**-0.5\n",
    "lossi = []"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(2.4525, grad_fn=<NllLossBackward0>)\n",
      "tensor(2.8024, grad_fn=<NllLossBackward0>)\n",
      "tensor(2.8380, grad_fn=<NllLossBackward0>)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for i in range(100_000):\n",
    "    loss = step(X_dev, Y_dev, lr)\n",
    "    lossi.append(torch.log10(torch.tensor([loss])).item())\n",
    "\n",
    "plt.plot(lossi)\n",
    "print(evaluate(X_dev,Y_dev))\n",
    "print(evaluate(X_tr,Y_tr))\n",
    "print(evaluate(X_tst,Y_tst))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[1;32m/Users/michalkonopka/github/my-grad/my_bengio.ipynb Cell 5\u001b[0m line \u001b[0;36m1\n\u001b[1;32m     <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=15'>16</a>\u001b[0m predicted \u001b[39m=\u001b[39m []\n\u001b[1;32m     <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=17'>18</a>\u001b[0m \u001b[39mwhile\u001b[39;00m \u001b[39mTrue\u001b[39;00m:\n\u001b[0;32m---> <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=18'>19</a>\u001b[0m    predicted_letter_index \u001b[39m=\u001b[39m predict_next_letter(letters_tensor)\n\u001b[1;32m     <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=19'>20</a>\u001b[0m    \u001b[39mif\u001b[39;00m predicted_letter_index \u001b[39m==\u001b[39m \u001b[39m0\u001b[39m:\n\u001b[1;32m     <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=20'>21</a>\u001b[0m       \u001b[39mbreak\u001b[39;00m\n",
      "\u001b[1;32m/Users/michalkonopka/github/my-grad/my_bengio.ipynb Cell 5\u001b[0m line \u001b[0;36m2\n\u001b[1;32m      <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=0'>1</a>\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mpredict_next_letter\u001b[39m(letters):\n\u001b[0;32m----> <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=1'>2</a>\u001b[0m     emb \u001b[39m=\u001b[39m C[letters]\n\u001b[1;32m      <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=2'>3</a>\u001b[0m     layer_1 \u001b[39m=\u001b[39m torch\u001b[39m.\u001b[39mtanh(emb\u001b[39m.\u001b[39mview(\u001b[39m-\u001b[39m\u001b[39m1\u001b[39m,\u001b[39m30\u001b[39m) \u001b[39m@\u001b[39m W1 \u001b[39m+\u001b[39m B1)\n\u001b[1;32m      <a href='vscode-notebook-cell:/Users/michalkonopka/github/my-grad/my_bengio.ipynb#W5sZmlsZQ%3D%3D?line=3'>4</a>\u001b[0m     layer_2 \u001b[39m=\u001b[39m layer_1 \u001b[39m@\u001b[39m W2 \u001b[39m+\u001b[39m B2\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "def predict_next_letter(letters):\n",
    "    emb = C[letters]\n",
    "    layer_1 = torch.tanh(emb.view(-1,30) @ W1 + B1)\n",
    "    layer_2 = layer_1 @ W2 + B2\n",
    "    prob_dist = torch.nn.functional.softmax(layer_2,dim=1)\n",
    "    next_letter = torch.multinomial(prob_dist, num_samples=1).item()\n",
    "    return next_letter\n",
    "\n",
    "def s_to_i(letters):\n",
    "   return list(map(lambda l: stoi[l],letters))\n",
    "\n",
    "\n",
    "for i in range(20):\n",
    "   letters = s_to_i(['.','.','.'])\n",
    "   letters_tensor = t(letters)\n",
    "   predicted = []\n",
    "\n",
    "   while True:\n",
    "      predicted_letter_index = predict_next_letter(letters_tensor)\n",
    "      if predicted_letter_index == 0:\n",
    "         break\n",
    "      predicted.append(predicted_letter_index)\n",
    "      letters = letters[1:] + [predict_next_letter]\n",
    "\n",
    "   name = \"\".join(list(map(lambda i: itos[i],predicted)))\n",
    "   print(f'{len(name)=}')\n",
    "      \n",
    "\n",
    "\n"
   ]
  }
 ],
 "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": "3.9.6"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}