-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathwesad_train_test.py
129 lines (112 loc) · 5.88 KB
/
wesad_train_test.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
from wesad import WESADTrSet, WESADTsSet
from avalanche.benchmarks.generators import dataset_benchmark
from avalanche.benchmarks.utils import AvalancheDataset
from avalanche.models import SimpleMLP
from avalanche.training.strategies import Naive, Cumulative, LwF, EWC, JointTraining, GEM, Replay
from torch.optim import Adam
from torch.nn import CrossEntropyLoss, MSELoss
from avalanche.evaluation.metrics import forgetting_metrics, accuracy_metrics, loss_metrics, timing_metrics, cpu_usage_metrics, confusion_matrix_metrics, disk_usage_metrics, gpu_usage_metrics
from avalanche.training.plugins import EvaluationPlugin, EarlyStoppingPlugin
from avalanche.logging import InteractiveLogger, TextLogger, TensorboardLogger
import pickle
import torch.nn as nn
import torch
import numpy as np
import sys
import time
class Classifier(nn.Module):
def __init__(self, in_dim, hidden_dim, out_dim, layers=1):
super().__init__()
self.rnn = nn.GRU(in_dim, hidden_dim, layers, batch_first=True)
self.clf = nn.Linear(hidden_dim, out_dim)
def forward(self, x):
# batch_size first
x, _ = self.rnn(x) # _ to ignore state
x = x[:, -1] # last timestep for classfication
return self.clf(x)
import warnings
warnings.filterwarnings("ignore")
def train_wesad(strat, i=""):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
scenario = dataset_benchmark(
[AvalancheDataset(WESADTrSet(pair=0), task_labels=1),
AvalancheDataset(WESADTrSet(pair=1), task_labels=2),
AvalancheDataset(WESADTrSet(pair=2), task_labels=3),
AvalancheDataset(WESADTrSet(pair=3), task_labels=4),
AvalancheDataset(WESADTrSet(pair=4), task_labels=5),
AvalancheDataset(WESADTrSet(pair=5), task_labels=6),
AvalancheDataset(WESADTrSet(pair=6), task_labels=7),
AvalancheDataset(WESADTrSet(pair=7), task_labels=8)],
[AvalancheDataset(WESADTsSet(), task_labels=0)]
)
tb_logger = TensorboardLogger()
text_logger = TextLogger(open('wesadlog.txt', 'a'))
int_logger = InteractiveLogger()
eval_plugin = EvaluationPlugin(
accuracy_metrics(minibatch=True, epoch=True, experience=True, stream=True),
loss_metrics(minibatch=True, epoch=True, experience=True, stream=True),
timing_metrics(epoch=True, epoch_running=True),
forgetting_metrics(experience=True, stream=True),
cpu_usage_metrics(experience=True),
gpu_usage_metrics(0, experience=True),
disk_usage_metrics(minibatch=True, epoch=True, experience=True, stream=True),
loggers=[text_logger]
)
es = EarlyStoppingPlugin(patience=25, val_stream_name="train_stream")
results = []
model = Classifier(in_dim=14, hidden_dim=18, out_dim=4, layers=2)
if (strat == "naive"):
print("Naive continual learning")
strategy = Naive(model, Adam(model.parameters(), lr=0.005, betas=(0.99, 0.99)), CrossEntropyLoss(), train_epochs=100, eval_every=1, plugins=[es], evaluator=eval_plugin, device=device)
elif (strat == "offline"):
print("Offline learning")
strategy = JointTraining(model, Adam(model.parameters(), lr=0.005, betas=(0.99, 0.99)), CrossEntropyLoss(), train_epochs=100, eval_every=1, plugins=[es], evaluator=eval_plugin, device=device)
elif (strat == "replay"):
print("Replay training")
strategy = Replay(model, Adam(model.parameters(), lr=0.005, betas=(0.99, 0.99)), CrossEntropyLoss(), train_epochs=100, eval_every=1, plugins=[es], evaluator=eval_plugin, device=device, mem_size=70, train_mb_size=70) #25% of WESAD
elif (strat == "cumulative"):
print("Cumulative continual learning")
strategy = Cumulative(model, Adam(model.parameters(), lr=0.005, betas=(0.99, 0.99)), CrossEntropyLoss(), train_epochs=100, eval_every=1, plugins=[es], evaluator=eval_plugin, device=device)
elif (strat == "lwf"):
print("LwF continual learning")
strategy = LwF(model, Adam(model.parameters(), lr=0.005, betas=(0.99, 0.99)), CrossEntropyLoss(), train_epochs=100, eval_every=1, plugins=[es], evaluator=eval_plugin, device=device, alpha=0.5, temperature=1)
elif (strat == "ewc"):
print("EWC continual learning")
torch.backends.cudnn.enabled = False
strategy = EWC(model, Adam(model.parameters(), lr=0.005, betas=(0.99, 0.99)), CrossEntropyLoss(), train_epochs=100, eval_every=1, plugins=[es], evaluator=eval_plugin, device=device, ewc_lambda=0.99)
elif (strat == "episodic"):
print("Episodic continual learning")
strategy = GEM(model, Adam(model.parameters(), lr=0.005, betas=(0.99, 0.99)), CrossEntropyLoss(), train_epochs=100, eval_every=1, plugins=[es], evaluator=eval_plugin, device=device, patterns_per_exp=70)
thisresults = []
print(i + ".")
start = time.time()
if strat == "offline":
res = strategy.train(scenario.train_stream)
r = strategy.eval(scenario.test_stream)
thisresults.append({"loss":r["Loss_Exp/eval_phase/test_stream/Task000/Exp000"],
"acc":(float(r["Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000"])*100),
"forg":r["StreamForgetting/eval_phase/test_stream"],
"all":r})
results.append({"strategy":strat,
"finalloss":r["Loss_Exp/eval_phase/test_stream/Task000/Exp000"],
"finalacc":r["Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000"],
"results":thisresults})
else:
for experience in scenario.train_stream:
res = strategy.train(experience)
r = strategy.eval(scenario.test_stream)
thisresults.append({"loss":r["Loss_Exp/eval_phase/test_stream/Task000/Exp000"],
"acc":(float(r["Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000"])*100),
"forg":r["StreamForgetting/eval_phase/test_stream"],
"all":r})
results.append({"strategy":strat,
"finalloss":r["Loss_Exp/eval_phase/test_stream/Task000/Exp000"],
"finalacc":r["Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000"],
"results":thisresults})
elapsed = time.time() - start
results.append({"time":elapsed})
with open("results/wesad_" + strat + "_results" + i + ".pkl", "wb") as outfile:
pickle.dump(results, outfile)
print("\t" + str(elapsed) + " seconds")
for i in range(5):
train_wesad(sys.argv[1].strip(), str(i))