-
Notifications
You must be signed in to change notification settings - Fork 0
/
mainCrossVal.py
186 lines (161 loc) · 8.35 KB
/
mainCrossVal.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
import torch
import os
from torch.utils.data import DataLoader
from torch import nn
from torchvision import transforms
from PIL import Image
import pandas as pd
from codec import EncoderCNN, DecoderRNN
from loaderTensor import ImageDataset
import numpy as np
import matplotlib.pyplot as plt
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
import numpy as np
# load data
root = "C:/pr_files/neural/"
#totalList = sorted(os.listdir(root+"images/"), key=len)
#lens = [int(len(totalList)*0.7), int(len(totalList)*0.1), len(totalList) - (int(len(totalList)*0.1) + int(len(totalList)*0.7))]
#train, test, val = torch.utils.data.dataset.random_split(totalList, lens)
tr = 0.05
val = 0.01
dataset_tr = ImageDataset(root_dir = "C:/pr_files/", trainPercent = tr, valPercent = val , flag = 'train', transform = transforms.Compose([transforms.RandomHorizontalFlip(), transforms.Resize((224,224),interpolation=Image.NEAREST),
transforms.ToTensor()]))
#dataset_test = ImageDataset(root_dir = root, trainPercent = tr, valPercent = val , flag = 'test',transform = transforms.Compose([transforms.Resize((224,224),interpolation=Image.NEAREST), transforms.ToTensor()]))
dataset_val = ImageDataset(root_dir = "C:/pr_files/", trainPercent = tr, valPercent = val , flag = 'validation',transform = transforms.Compose([transforms.Resize((224,224),interpolation=Image.NEAREST),
transforms.ToTensor(), transforms.Normalize(mean = [0.489, 0.456, 0.406], std = [0.229, 0.224,0.225])]))
wordC = pd.read_hdf(root + "eee443_project_dataset_train.h5", 'word_code')
wordC = wordC.to_dict('split')
wordDict = dict(zip(wordC['data'][0], wordC['columns']))
dataloader_val = DataLoader(dataset_val, batch_size=50, shuffle=True)
valBatchSize = 2**np.arange(2,8)
avrValLoss = []
#cross validate across decoder hidden layer size
for j in valBatchSize:
dataloader_train = DataLoader(dataset_tr, batch_size=int(j), shuffle=True)
print('Batch Size', j)
# Initializations for training
losses = list()
val_losses = list()
decoder = DecoderRNN(embed_size=300, hidden_size=256, vocab_size=1004).to(device)
#encoder = EncoderCNN(300).to(device)
# encoder.fine_tune()
criterion = nn.CrossEntropyLoss().to(device)
vocab_size = 1004
# encoder_lr = 0.008
# encoder_optimizer = torch.optim.Adam(params=filter(lambda p: p.requires_grad, encoder.parameters()),
# lr=encoder_lr)
decoder_lr = 0.005
decoder_optimizer = torch.optim.Adam(params=filter(lambda p: p.requires_grad, decoder.parameters()),
lr=decoder_lr)
# Training starts
for epoch in range(1, 3):
print("Epoch:",epoch)
for i, data in enumerate(dataloader_train):
# zero the gradients
decoder.zero_grad()
# encoder.zero_grad()
# set decoder and encoder into train mode
# encoder.train()
decoder.train()
# Obtain the batch.
images = data["image"]
captions = data["caption"]
# make the captions for targets and teacher forcer
captions_target = captions[:, 1:].long().to(device)
captions_train = captions[:, :captions.shape[1]-1].long().to(device)
# Move batch of images and captions to GPU if CUDA is available.
images = images.to(device)
# Pass the inputs through the CNN-RNN model.
outputs = decoder(images, captions_train)
# Calculate the batch loss
trainLoss = criterion(outputs.view(-1, vocab_size), captions_target.contiguous().view(-1))
# Backward pass
trainLoss.backward()
# Update the parameters in the optimizer
# encoder_optimizer.step()
decoder_optimizer.step()
if i % 200 == 0:
print('Batch ', i)
# # - - - Validate - - -
# # turn the evaluation mode on
# if i % 50 == 0:
# with torch.no_grad():
#
# # set the evaluation mode
# decoder.eval()
#
# # get the validation images and captions
# dataVal = next(iter(dataloader_val))
# val_images = dataVal["image"]
# img = val_images[0]
# val_captions = dataVal["caption"]
# cap = val_captions[0]
#
# caption = cap.cpu().detach().numpy()
# captionStr = [wordDict[i] for i in caption]
# print(captionStr)
# # define the captions
# captions_target = val_captions[:, 1:].long().to(device)
# captions_train = val_captions[:, :-1].long().to(device)
#
# # Move batch of images and captions to GPU if CUDA is available.
# val_images = val_images.to(device)
#
# # Pass the inputs through the CNN-RNN model.
## val_features = encoder(val_images)
# val_outputs = decoder(val_images, captions_train)
#
# captionOut = val_outputs[0].cpu().detach().numpy().argmax(axis = 1)
# captionOutStr = [wordDict[i] for i in captionOut]
# print("\n", captionOutStr)
# # Calculate the batch loss.
# val_loss = criterion(val_outputs.view(-1, vocab_size), captions_target.contiguous().view(-1))
# print("Loss of ", i, "th batch: ", val_loss.item(), sep="")
#
# val_losses.append(val_loss.item())
# losses.append(trainLoss.item())
# np.save("train_lossesWOEncoder", np.array(losses))
# np.save("val_lossesWOEncoder", np.array(val_losses))
# append the validation loss and training loss
print("\nSaving the model")
# time_stamp = str(int(__import__("time").time()))
torch.save(decoder.state_dict(), "./models/decoderCrossValBatchSize"+ str(j) + ".pth")
# torch.save(encoder.state_dict(), "./models/encoderWOEncoder"+ time_stamp + ".pth")
for i, valData in enumerate(dataloader_val):
# - - - Validate - - -
# turn the evaluation mode on
with torch.no_grad():
# set the evaluation mode
decoder.eval()
# get the validation images and captions
dataVal = valData
val_images = dataVal["image"]
img = val_images[0]
val_captions = dataVal["caption"]
cap = val_captions[0]
caption = cap.cpu().detach().numpy()
# captionStr = [wordDict[i] for i in caption]
# print(captionStr)
# define the captions
captions_target = val_captions[:, 1:].long().to(device)
captions_train = val_captions[:, :-1].long().to(device)
# Move batch of images and captions to GPU if CUDA is available.
val_images = val_images.to(device)
# Pass the inputs through the CNN-RNN model.
# val_features = encoder(val_images)
val_outputs = decoder(val_images, captions_train)
captionOut = val_outputs[0].cpu().detach().numpy().argmax(axis = 1)
captionOutStr = [wordDict[i] for i in captionOut]
print("\n", captionOutStr)
# Calculate the batch loss.
val_loss = criterion(val_outputs.view(-1, vocab_size), captions_target.contiguous().view(-1))
print("Loss of ", i, "th batch: ", val_loss.item(), sep="")
val_losses.append(val_loss.item())
#losses.append(trainLoss.item())
avrValLoss.append( sum(val_losses)/len(val_losses))
plt.plot(valBatchSize,avrValLoss)
plt.xlabel('Batch Size')
plt.ylabel('Cross Entropy Loss')
plt.title('Validation set Cross Entropy Loss \n with different Batch Sizes')
#np.save("train_lossesWOEncoder", np.array(losses))
np.save("crossVal_batchSize", np.array(val_losses))