-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathMain_TwoSampleStatsNN_StandardSplit.py
183 lines (146 loc) · 8.05 KB
/
Main_TwoSampleStatsNN_StandardSplit.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
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
from __future__ import print_function
import os
import pickle
import numpy as np
import torch
import torch.optim as optim
import random
from DataUtils import ExtractOneHoldStandardSplit
from Models import ChemCam_CNN, MyAIDataset, TwoSampletrain, test
from Options import *
from sklearn import preprocessing
def main():
# Training settings
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
use_mps = not args.no_mps and torch.backends.mps.is_available()
# set random seed
np.random.seed(args.seed)
random.seed(args.seed)
torch.manual_seed(args.seed)
if use_cuda:
device = torch.device("cuda")
elif use_mps:
device = torch.device("mps")
else:
device = torch.device("cpu")
train_kwargs = {'batch_size': args.batch_size}
test_kwargs = {'batch_size': args.test_batch_size}
if use_cuda:
cuda_kwargs1 = {'num_workers': 1,
'pin_memory': True,
'shuffle': True}
cuda_kwargs2 = {'num_workers': 1,
'pin_memory': True,
'shuffle': False}
train_kwargs.update(cuda_kwargs1)
test_kwargs.update(cuda_kwargs2)
# Set up directory
cwd = os.getcwd()
if args.LossType2 == 'fdiv':
ResultDir = cwd + '/%s/Results/fdiv/Epochs%dBatch%dLambda%.3fLR%.3fWeight%.6fDelta%.6f/'%(args.DataType, args.epochs, args.batch_size, args.Lambda, args.lr, args.weight, args.delta)
elif args.LossType2 == 'L2_fdiv':
ResultDir = cwd + '/%s/Results/L2_fdiv/Epochs%dBatch%dLambda%.3fLR%.3fWeight%.6fDelta%.6f_l2strength%.6f/'%(args.DataType, args.epochs, args.batch_size, args.Lambda, args.lr, args.weight, args.delta, args.l2delta)
elif args.LossType2 == 'L1_fdiv':
ResultDir = cwd + '/%s/Results/L1_fdiv/Epochs%dBatch%dLambda%.3fLR%.3fWeight%.6fDelta%.6f_l1strength%.6f/'%(args.DataType, args.epochs, args.batch_size, args.Lambda, args.lr, args.weight, args.delta, args.l1delta)
elif args.LossType2 == 'Dropout_fdiv':
ResultDir = cwd + '/%s/Results/Dropout_fdiv/Epochs%dBatch%dLambda%.3fLR%.3fWeight%.6fDelta%.6f_dropoutstrength%.6f/'%(args.DataType, args.epochs, args.batch_size, args.Lambda, args.lr, args.weight, args.delta, args.Dropout_prob)
# Set the saving directories
ModelSaveDir = ResultDir + 'Models/'
StatsSaveDir = ResultDir + 'Stats/'
if not os.path.exists(ResultDir):
os.makedirs(ResultDir)
if not os.path.exists(ModelSaveDir):
os.makedirs(ModelSaveDir)
if not os.path.exists(StatsSaveDir):
os.makedirs(StatsSaveDir)
# Construct variables and acquire the training, validation and testing splits of the original data
SampleSplitDir = cwd + '/%s/SampleSplits/'%(args.DataType)
with open(SampleSplitDir + 'Seed%d_TrSampleName.pickle'%args.seed, 'rb') as file:
TrSampleName = pickle.load(file)
with open(SampleSplitDir + 'Seed%d_ValSampleName.pickle'%args.seed, 'rb') as file:
ValSampleName = pickle.load(file)
with open(SampleSplitDir + 'Seed%d_TeSampleName.pickle'%args.seed, 'rb') as file:
TeSampleName = pickle.load(file)
# Extract train, validation and test sets
TrFeat, ValFeat, TeFeat, \
TrLabel, ValLabel, TeLabel, \
ValSampleName, TeSampleName, ValShotCount, TeShotCount = ExtractOneHoldStandardSplit(args, TrSampleName, ValSampleName, TeSampleName)
# Preprocess the data
TrFeat = preprocessing.normalize(TrFeat) # remove outlier intensities
ValFeat = preprocessing.normalize(ValFeat)
TeFeat = preprocessing.normalize(TeFeat)
# Label scalling
TrLabel = TrLabel/100; ValLabel = ValLabel/100; TeLabel = TeLabel/100
# Construct dictionaries that save the predictions
ValPredictLabelDic = {}; TePredictLabelDic = {}
# normalize training targets
scaler = preprocessing.StandardScaler().fit(TrLabel)
TrLabel = scaler.transform(TrLabel)
# Set variables
ValErrEpoches1 = np.ones((args.epochs));
ValErrEpoches2 = np.ones((args.epochs));
# Set pytorch datasets
TrDataSet = MyAIDataset(TrFeat, TrLabel)
ValDataSet = MyAIDataset(ValFeat, ValLabel)
TeDataSet = MyAIDataset(TeFeat, TeLabel)
# Convert data to pytorch datasets
train_loader = torch.utils.data.DataLoader(TrDataSet, **train_kwargs)
val_loader = torch.utils.data.DataLoader(ValDataSet, **test_kwargs)
te_loader = torch.utils.data.DataLoader(TeDataSet, **test_kwargs)
# Build models
if args.DataType == 'PDS_ChemCam':
if args.LossType2 == 'Dropout_fdiv':
predmodel = ChemCam_CNN(args.Dropout_prob).to(device) # one-D CNN
else:
predmodel = ChemCam_CNN().to(device) # one-D CNN
elif args.DataType == 'PDS_SuperCam':
pass
if args.LossType2 == 'L2_fdiv':
pred_optimizer = optim.Adadelta(predmodel.parameters(), lr=args.lr, weight_decay=args.l2delta)
else:
pred_optimizer = optim.Adadelta(predmodel.parameters(), lr=args.lr)
# Set variable to save training details
MinErr1 = np.inf; MinErr2 = np.inf;
Fdivtrain2 = TwoSampletrain(args, predmodel, device, train_loader, val_loader, pred_optimizer)
for epoch in range(1, args.epochs + 1):
Fdivtrain2.train(epoch)
Pred, Err1, Fdiv = test(args, predmodel, device, val_loader, scaler) # output prediction for the validation set, validation mse and f-divergence
ValErrEpoches1[int(epoch - 1)] = Err1
ValErrEpoches2[int(epoch - 1)] = Fdiv
if Err1 < MinErr1:
ValPred = Pred; MinErr1 = Err1
if args.save_model:
torch.save(predmodel.state_dict(), ModelSaveDir+ "MSEnetwork_StandardSplit%d.pt"%args.seed)
if Fdiv < MinErr2:
MinErr2 = Fdiv
if args.save_model:
torch.save(predmodel.state_dict(), ModelSaveDir+ "fdivnetwork_StandardSplit%d.pt"%args.seed)
# Evaluate model on the test set
predmodel.load_state_dict(torch.load(ModelSaveDir + 'MSEnetwork_StandardSplit%d.pt'%args.seed))
TePred, TeErr1, TeFdiv = test(args, predmodel, device, te_loader, scaler)
# print results
if args.LossType2 == 'fdiv':
print('DataType:%s, Loss:%s, w:%.5f, gamma:%.5f, Fold:%d, Val MSE: %.5f, Test MSE: %.5f'%(args.DataType, args.LossType2, args.weight, args.delta, args.seed, MinErr1, TeErr1))
elif args.LossType2 == 'L2_fdiv':
print('DataType:%s, Loss:%s, l2 strength:%.5f, w:%.5f, gamma:%.5f, Fold:%d, Val MSE: %.5f, Test MSE: %.5f'%(args.DataType, args.LossType2, args.l2delta, args.weight, args.delta, args.seed, MinErr1, TeErr1))
elif args.LossType2 == 'L1_fdiv':
print('DataType:%s, Loss:%s, l1 strength:%.5f, w:%.5f, gamma:%.5f, Fold:%d, Val MSE: %.5f, Test MSE: %.5f'%(args.DataType, args.LossType2, args.l1delta, args.weight, args.delta, args.seed, MinErr1, TeErr1))
elif args.LossType2 == 'Dropout_fdiv':
print('DataType:%s, Loss:%s, dropout strength:%.5f, w:%.5f, gamma:%.5f, Fold:%d, Val MSE: %.5f, Test MSE: %.5f'%(args.DataType, args.LossType2, args.Dropout_prob, args.weight, args.delta, args.seed, MinErr1, TeErr1))
# save results to dictionaries
for j, sn in enumerate(ValSampleName):
ValPredictLabelDic[sn] = ValPred[j]
for j, sn in enumerate(TeSampleName):
TePredictLabelDic[sn] = TePred[j]
# Save predictions
with open(StatsSaveDir + 'Seed%d_TePredictLabelDic_StandardSplit.pickle'%(args.seed), 'wb') as file:
pickle.dump(TePredictLabelDic, file, protocol=pickle.HIGHEST_PROTOCOL)
with open(StatsSaveDir + 'Seed%d_ValPredictLabelDic_StandardSplit.pickle'%(args.seed), 'wb') as file:
pickle.dump(ValPredictLabelDic, file, protocol=pickle.HIGHEST_PROTOCOL)
with open(StatsSaveDir + 'Seed%d_TeSampleIdx_StandardSplit.pickle'%(args.seed), 'wb') as file:
pickle.dump(TeSampleName, file, protocol=pickle.HIGHEST_PROTOCOL)
with open(StatsSaveDir + 'Seed%d_ValSampleIdx_StandardSplit.pickle'%(args.seed), 'wb') as file:
pickle.dump(ValSampleName, file, protocol=pickle.HIGHEST_PROTOCOL)
if __name__ == '__main__':
main()