forked from brjathu/RPSnet
-
Notifications
You must be signed in to change notification settings - Fork 0
/
mnist.py
329 lines (247 loc) · 10.5 KB
/
mnist.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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
'''
RPS network training on CIFAR100
Copyright (c) Jathushan Rajasegaran, 2019
'''
from __future__ import print_function
import argparse
import os
import shutil
import time
import random
import pickle
import torch
import pdb
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.optim as optim
import torch.utils.data as data
import torchvision.transforms as transforms
import torchvision.datasets as datasets
from utils import Bar, Logger, AverageMeter, accuracy, mkdir_p, savefig
import numpy as np
import copy
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
from torch.autograd import gradcheck
import sys
import random
from torch.utils.data import Dataset, TensorDataset
from rps_net import RPS_net_mlp, RPS_net_cifar
from learner import Learner
from util import *
from cifar_dataset import CIFAR100
class args:
epochs = 10
checkpoint = "results/mnist/RPS_net_minst"
savepoint = "results/mnist/pathnet_mnist"
dataset = "MNIST"
num_class = 10
class_per_task = 2
M = 8
L = 9
N = 1
lr = 0.001
train_batch = 128
test_batch = 128
workers = 16
resume = False
arch = "res-18"
start_epoch = 0
evaluate = False
sess = 0
test_case = 0
schedule = [6, 8, 16]
gamma = 0.5
rigidness_coff = 2.5
jump = 1
state = {key:value for key, value in args.__dict__.items() if not key.startswith('__') and not callable(key)}
print(state)
memory = 4400
# Use CUDA
use_cuda = torch.cuda.is_available()
seed = random.randint(1, 10000)
random.seed(seed)
torch.manual_seed(seed)
if use_cuda:
torch.cuda.manual_seed_all(seed)
def load_mnist():
from keras.datasets import mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(-1, 784).astype('float32') / 255.
x_test = x_test.reshape(-1, 784).astype('float32') / 255.
return (x_train, y_train), (x_test, y_test)
def load_svhn():
from scipy import io as spio
from keras.utils import to_categorical
import numpy as np
svhn = spio.loadmat("train_32x32.mat")
x_train = np.einsum('ijkl->lijk', svhn["X"]).astype(np.float32) / 255.
y_train = (svhn["y"] - 1)
svhn_test = spio.loadmat("test_32x32.mat")
x_test = np.einsum('ijkl->lijk', svhn_test["X"]).astype(np.float32) / 255.
y_test = (svhn_test["y"] - 1)
x_train = np.transpose(x_train, [0,3,1,2])
x_test = np.transpose(x_test, [0,3,1,2])
y_train = np.reshape(y_train, (-1))
y_test = np.reshape(y_test, (-1))
return (x_train, y_train), (x_test, y_test)
def load_cifar10():
from keras.datasets import cifar10
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train = x_train.reshape(-1, 3, 32, 32).astype('float32') / 255.
x_test = x_test.reshape(-1, 3, 32, 32).astype('float32') / 255.
y_train = np.reshape(y_train, (-1))
y_test = np.reshape(y_test, (-1))
return (x_train, y_train), (x_test, y_test)
class CustomTensorDataset(Dataset):
"""TensorDataset with support of transforms.
"""
def __init__(self, tensors, transform=None):
assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors)
self.tensors = tensors
self.transform = transform
def __getitem__(self, index):
x = self.tensors[0][index]
if self.transform:
x = self.transform(x)
y = self.tensors[1][index]
return x, y
def __len__(self):
return self.tensors[0].size(0)
def main():
if not os.path.isdir(args.checkpoint):
mkdir_p(args.checkpoint)
if not os.path.isdir("models/mnist/"+args.checkpoint.split("/")[-1]):
mkdir_p("models/mnist/"+args.checkpoint.split("/")[-1])
args.savepoint = "models/mnist/"+args.checkpoint.split("/")[-1]
model = RPS_net_mlp(args).cuda()
# model = RPS_net_cifar(args).cuda() #for SVHN and CIFAR10
print(' Total params: %.2fM' % (sum(p.numel() for p in model.parameters())/1000000.0))
start_sess = int(sys.argv[2])
test_case = sys.argv[1]
args.test_case = test_case
# (x_train, y_train), (x_test, y_test) = load_svhn()
# (x_train, y_train), (x_test, y_test) = load_cifar10()
(x_train, y_train), (x_test, y_test) = load_mnist()
for ses in range(start_sess, start_sess+1):
if(ses==0):
path = get_path(args.L,args.M,args.N)*0
path[:,0] = 1
fixed_path = get_path(args.L,args.M,args.N)*0
train_path = path.copy()
infer_path = path.copy()
else:
load_test_case = get_best_model(ses-1, args.checkpoint)
if(ses%args.jump==0): #get a new path
fixed_path = np.load(args.checkpoint+"/fixed_path_"+str(ses-1)+"_"+str(load_test_case)+".npy")
path = get_path(args.L,args.M,args.N)
train_path = get_path(args.L,args.M,args.N)*0
else:
if((ses//args.jump)*2==0):
fixed_path = get_path(args.L,args.M,args.N)*0
else:
load_test_case_x = get_best_model((ses//args.jump)*args.jump-1, args.checkpoint)
fixed_path = np.load(args.checkpoint+"/fixed_path_"+str((ses//args.jump)*args.jump-1)+"_"+str(load_test_case_x)+".npy")
path = np.load(args.checkpoint+"/path_"+str(ses-1)+"_"+str(load_test_case)+".npy")
train_path = get_path(args.L,args.M,args.N)*0
infer_path = get_path(args.L,args.M,args.N)*0
for j in range(args.L):
for i in range(args.M):
if(fixed_path[j,i]==0 and path[j,i]==1):
train_path[j,i]=1
if(fixed_path[j,i]==1 or path[j,i]==1):
infer_path[j,i]=1
np.save(args.checkpoint+"/path_"+str(ses)+"_"+str(test_case)+".npy", path)
if(ses==0):
fixed_path_x = path.copy()
else:
fixed_path_x = fixed_path.copy()
for j in range(args.L):
for i in range(args.M):
if(fixed_path_x[j,i]==0 and path[j,i]==1):
fixed_path_x[j,i]=1
np.save(args.checkpoint+"/fixed_path_"+str(ses)+"_"+str(test_case)+".npy", fixed_path_x)
print('Starting with session {:d}'.format(ses))
print('test case : ' + str(test_case))
print('#################################################################################')
print("path\n",path)
print("fixed_path\n",fixed_path)
print("train_path\n", train_path)
print("infer_path\n", infer_path)
ids_train = []
for j in range((ses*args.class_per_task), (ses+1)*args.class_per_task):
ids_train.append(np.where(y_train==j)[0])
ids_test = []
for j in range((ses+1)*args.class_per_task):
ids_test.append(np.where(y_test==j)[0])
ids_train = flatten_list(ids_train)
ids_test = flatten_list(ids_test)
if(ses>0):
ids_exp = []
for j in range((ses)*args.class_per_task):
ex_id =np.where(y_train==j)[0]
sample_per_class = memory//(ses*args.class_per_task)
if(len(ex_id)>sample_per_class):
ids_exp.append(ex_id[0:sample_per_class])
else:
ids_exp.append(ex_id)
ids_exp = np.tile(flatten_list(ids_exp),10)
train_data = np.vstack([x_train[ids_train], x_train[ids_exp]])
train_label = np.vstack([np.reshape(y_train[ids_train],(-1,1)), np.reshape(y_train[ids_exp],(-1,1))])
train_label = flatten_list(train_label)
else:
ids_exp = []
train_data = x_train[ids_train]
train_label = y_train[ids_train]
test_data = x_test[ids_test]
test_label = y_test[ids_test]
import torch.utils.data as utils
args.sess = ses
transform_train = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize(32),
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.RandomRotation(10),
transforms.ToTensor(),
])
transform_test = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize(32),
transforms.ToTensor(),
])
train_dataset = utils.TensorDataset(torch.from_numpy(train_data).float(), torch.from_numpy(train_label).long()) # create your datset
train_dataset2 = CustomTensorDataset((torch.tensor(train_data), torch.tensor(train_label).long()), transform=transform_train)
train_loader = utils.DataLoader(train_dataset, batch_size=args.train_batch, shuffle=True) # create your data
test_dataset = utils.TensorDataset(torch.from_numpy(test_data).float(), torch.from_numpy(test_label).long()) # create your datset
test_dataset2 = CustomTensorDataset((torch.tensor(test_data), torch.tensor(test_label).long()), transform=transform_test)
test_loader = utils.DataLoader(test_dataset, batch_size=args.test_batch)
main_learner=Learner(model=model,args=args,trainloader=train_loader,
testloader=test_loader,old_model=copy.deepcopy(model),
use_cuda=use_cuda, path=path,
fixed_path=fixed_path, train_path=train_path, infer_path=infer_path)
main_learner.learn()
if(ses==0):
fixed_path = path.copy()
else:
for j in range(args.L):
for i in range(args.M):
if(fixed_path[j,i]==0 and path[j,i]==1):
fixed_path[j,i]=1
np.save(args.checkpoint+"/fixed_path_"+str(ses)+"_"+str(test_case)+".npy", fixed_path)
best_model = get_best_model(ses, args.checkpoint)
cfmat = main_learner.get_confusion_matrix(infer_path)
np.save(args.checkpoint+"/confusion_matrix_"+str(ses)+"_"+str(test_case)+".npy", cfmat)
print('done with session {:d}'.format(ses))
print('#################################################################################')
while(1):
if(is_all_done(ses, args.epochs, args.checkpoint)):
break
else:
time.sleep(10)
if __name__ == '__main__':
main()