-
Notifications
You must be signed in to change notification settings - Fork 0
/
datahandler.py
475 lines (393 loc) · 16 KB
/
datahandler.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
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
import os
from torchvision import transforms, datasets, models
import torch
from torch import optim, cuda
from torch.utils.data import DataLoader, sampler
import torch.nn as nn
import numpy as np
import pandas as pd
import os
from PIL import Image
from torchsummary import summary
from timeit import default_timer as timer
import matplotlib.pyplot as plt
from inceptionv4 import inceptionv4
n_classes=2
train_on_gpu=cuda.is_available()
if train_on_gpu:
gpu_count = cuda.device_count()
print(f'{gpu_count} gpus detected.')
if gpu_count > 1:
multi_gpu = True
else:
multi_gpu = False
def process_image(image_path):
"""Process an image path into a PyTorch tensor"""
image = Image.open(image_path)
# Resize
img = image.resize((256, 256))
# Center crop
width = 256
height = 256
new_width = 224
new_height = 224
left = (width - new_width) / 2
top = (height - new_height) / 2
right = (width + new_width) / 2
bottom = (height + new_height) / 2
img = img.crop((left, top, right, bottom))
# Convert to numpy, transpose color dimension and normalize
img = np.array(img).transpose((2, 0, 1)) / 256
# Standardization
means = np.array([0.485, 0.456, 0.406]).reshape((3, 1, 1))
stds = np.array([0.229, 0.224, 0.225]).reshape((3, 1, 1))
img = img - means
img = img / stds
img_tensor = torch.Tensor(img)
return img_tensor
def imshow_tensor(image, ax=None, title=None):
"""Imshow for Tensor."""
if ax is None:
fig, ax = plt.subplots()
# Set the color channel as the third dimension
image = image.numpy().transpose((1, 2, 0))
# Reverse the preprocessing steps
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
image = std * image + mean
# Clip the image pixel values
image = np.clip(image, 0, 1)
ax.imshow(image)
plt.axis('off')
return ax, image
def train(model,
criterion,
optimizer,
train_loader,
valid_loader,
save_file_name,
max_epochs_stop=3,
n_epochs=20,
print_every=2,history=[]):
"""Train a PyTorch Model
Params
--------
model (PyTorch model): cnn to train
criterion (PyTorch loss): objective to minimize
optimizer (PyTorch optimizier): optimizer to compute gradients of model parameters
train_loader (PyTorch dataloader): training dataloader to iterate through
valid_loader (PyTorch dataloader): validation dataloader used for early stopping
save_file_name (str ending in '.pt'): file path to save the model state dict
max_epochs_stop (int): maximum number of epochs with no improvement in validation loss for early stopping
n_epochs (int): maximum number of training epochs
print_every (int): frequency of epochs to print training stats
Returns
--------
model (PyTorch model): trained cnn with best weights
history (DataFrame): history of train and validation loss and accuracy
"""
# Early stopping intialization
epochs_no_improve = 0
valid_loss_min = np.Inf
valid_max_acc = 0
#history = []
# Number of epochs already trained (if using loaded in model weights)
try:
print(f'Model has been trained for: {model.epochs} epochs.\n')
except:
model.epochs = 0
print(f'Starting Training from Scratch.\n')
overall_start = timer()
# Main loop
for epoch in range(n_epochs):
# keep track of training and validation loss each epoch
train_loss = 0.0
valid_loss = 0.0
train_acc = 0
valid_acc = 0
# Set to training
model.train()
start = timer()
# Training loop
for ii, (data, target) in enumerate(train_loader):
# Tensors to gpu
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# Clear gradients
optimizer.zero_grad()
# Predicted outputs are log probabilities
output = model(data)
# Loss and backpropagation of gradients
loss = criterion(output, target)
loss.backward()
# Update the parameters
optimizer.step()
# Track train loss by multiplying average loss by number of examples in batch
train_loss += loss.item() * data.size(0)
# Calculate accuracy by finding max log probability
_, pred = torch.max(output, dim=1)
correct_tensor = pred.eq(target.data.view_as(pred))
# Need to convert correct tensor from int to float to average
accuracy = torch.mean(correct_tensor.type(torch.FloatTensor))
# Multiply average accuracy times the number of examples in batch
train_acc += accuracy.item() * data.size(0)
# Track training progress
print(
f'Epoch: {epoch}\t{100 * (ii + 1) / len(train_loader):.2f}% complete. {timer() - start:.2f} seconds elapsed in epoch.',
end='\r')
# After training loops ends, start validation
else:
model.epochs += 1
# Don't need to keep track of gradients
with torch.no_grad():
# Set to evaluation mode
model.eval()
# Validation loop
for data, target in valid_loader:
# Tensors to gpu
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# Forward pass
output = model(data)
# Validation loss
loss = criterion(output, target)
# Multiply average loss times the number of examples in batch
valid_loss += loss.item() * data.size(0)
# Calculate validation accuracy
_, pred = torch.max(output, dim=1)
correct_tensor = pred.eq(target.data.view_as(pred))
accuracy = torch.mean(
correct_tensor.type(torch.FloatTensor))
# Multiply average accuracy times the number of examples
valid_acc += accuracy.item() * data.size(0)
# Calculate average losses
train_loss = train_loss / len(train_loader.dataset)
valid_loss = valid_loss / len(valid_loader.dataset)
# Calculate average accuracy
train_acc = train_acc / len(train_loader.dataset)
valid_acc = valid_acc / len(valid_loader.dataset)
history.append([train_loss, valid_loss, train_acc, valid_acc])
# Print training and validation results
if (epoch + 1) % print_every == 0:
print(
f'\nEpoch: {epoch} \tTraining Loss: {train_loss:.4f} \tValidation Loss: {valid_loss:.4f}'
)
print(
f'\t\tTraining Accuracy: {100 * train_acc:.2f}%\t Validation Accuracy: {100 * valid_acc:.2f}%'
)
#save_checkpoint(model,save_file_name.split('.')[0]+(f"-epoch_{model.epochs}.pth"),history)
# Save the model if validation loss decreases
save_checkpoint(model,save_file_name.split('.')[0]+f"-epoch_{model.epochs}.pth",optimizer,history)
if valid_loss < valid_loss_min:
# Save model
torch.save(model.state_dict(), save_file_name.split('.')[0]+f"-best_.pth")
#save_checkpoint(model,save_file_name,optimizer,history)
# Track improvement
epochs_no_improve = 0
valid_loss_min = valid_loss
valid_best_acc = valid_acc
best_epoch = epoch
# Otherwise increment count of epochs with no improvement
else:
epochs_no_improve += 1
# Trigger early stopping
if epochs_no_improve >= max_epochs_stop:
print(
f'\nEarly Stopping! Total epochs: {epoch}. Best epoch: {best_epoch} with loss: {valid_loss_min:.2f} and acc: {100 * valid_acc:.2f}%'
)
total_time = timer() - overall_start
print(
f'{total_time:.2f} total seconds elapsed. {total_time / (epoch+1):.2f} seconds per epoch.'
)
# Load the best state dict
#checkpoint = torch.load(path)
model.load_state_dict(torch.load(save_file_name.split('.')[0]+f"-best_.pth"))
# Attach the optimizer
#model.optimizer = optimizer
# Format history
history = pd.DataFrame(
history,
columns=[
'train_loss', 'valid_loss', 'train_acc',
'valid_acc'
])
return model, history
# Attach the optimizer
model.optimizer = optimizer
# Record overall time and print out stats
total_time = timer() - overall_start
print(
f'\nBest epoch: {best_epoch} with loss: {valid_loss_min:.2f} and acc: {100 * valid_acc:.2f}%'
)
print(
f'{total_time:.2f} total seconds elapsed. {total_time / (epoch):.2f} seconds per epoch.'
)
# Format history
history = pd.DataFrame(
history,
columns=['train_loss', 'valid_loss', 'train_acc', 'valid_acc'])
return model, history
def load_checkpoint(path):
"""Load a PyTorch model checkpoint
Params
--------
path (str): saved model checkpoint. Must start with `model_name-` and end in '.pth'
Returns
--------
None, save the `model` to `path`
"""
# Get the model name
model_name = path.split('-')[0]
assert (model_name in ['vgg16', 'resnet34','densenet121','inceptionv4']), "Path must have the correct model name"
checkpoint = torch.load(path)
if model_name == 'vgg16':
model = models.vgg16(pretrained=True)
# Make sure to set parameters as not trainable
for param in model.parameters():
param.requires_grad = False
model.classifier = checkpoint['classifier']
elif model_name == 'resnet34':
model=models.resnet34(pretrained=True)
# Freeze model weights
for parm in model.parameters():
parm.requires_grad=False
model.fc = nn.Sequential(
nn.Linear(512, 128),
nn.ReLU(inplace=True),
nn.Linear(128, 2))
model = model.cuda()
elif model_name == 'densenet121':
model = models.densenet121(pretrained=True)
# Make sure to set parameters as not trainable
for param in model.parameters():
param.requires_grad = False
num_ftrs = model.classifier.in_features
model.classifier=nn.Linear(num_ftrs,2)
model = model.cuda()
elif model_name=='inceptionv4':
model=inceptionv4(pretrained='imagenet')
for param in model.parameters():
param.requires_grad=False
num_ftrs = model.last_linear.in_features
model.last_linear = nn.Sequential(
nn.Linear(num_ftrs, 128),
nn.ReLU(inplace=True),
nn.Linear(128, 2))
model = model.cuda()
#model.optimizer=optim.Adam(model.parameters())
optimizer = checkpoint['optimizer']
model.load_state_dict(checkpoint['model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
model.epochs = checkpoint['epoch']
model.class_to_idx = checkpoint['class_to_idx']
model.idx_to_class = checkpoint['idx_to_class']
model.eval()
total_params = sum(p.numel() for p in model.parameters())
print(f'{total_params:,} total parameters.')
total_trainable_params = sum(
p.numel() for p in model.parameters() if p.requires_grad)
print(f'{total_trainable_params:,} total gradient parameters.')
# Move to gpu
if multi_gpu:
model = nn.DataParallel(model)
if train_on_gpu:
model = model.to('cuda')
return model, optimizer,checkpoint['history']
def save_checkpoint(model, path,optimizer,history):
"""Save a PyTorch model checkpoint
Params
--------
model (PyTorch model): model to save
path (str): location to save model. Must start with `model_name-` and end in '.pth'
Returns
--------
None, save the `model` to `path`
"""
model_name = path.split('-')[0]
assert (model_name in ['vgg16','resnet34','densenet121','inceptionv4'
]), "Path must have the correct model name"
# Extract the final classifier and the state dictionary
torch.save({
'class_to_idx': model.class_to_idx,
'idx_to_class': model.idx_to_class,
'epoch': model.epochs,
'model_state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'optimizer' : optimizer,
'history': history,
}, path)
# Add the optimizer
#checkpoint['optimizer'] = model.optimizer
#checkpoint['optimizer_state_dict'] = model.optimizer.state_dict()
# Save the data to the path
#torch.save(checkpoint, path)
def display_prediction(image_path, model, topk):
"""Display image and preditions from model"""
# Get predictions
img, ps, classes, y_obs = predict(image_path, model, topk)
print(predict(image_path, model, topk))
# Convert results to dataframe for plotting
return
'''
result = pd.DataFrame({'p': ps}, index=classes)
# Show the image
plt.figure(figsize=(16, 5))
ax = plt.subplot(1, 2, 1)
ax, img = imshow_tensor(img, ax=ax)
# Set title to be the actual class
ax.set_title(y_obs, size=20)
ax = plt.subplot(1, 2, 2)
# Plot a bar plot of predictions
result.sort_values('p')['p'].plot.barh(color='blue', edgecolor='k', ax=ax)
plt.xlabel('Predicted Probability')
plt.tight_layout()
'''
def predict(image_path, model, topk=2,splitter='\\'):
"""Make a prediction for an image using a trained model
Params
--------
image_path (str): filename of the image
model (PyTorch model): trained model for inference
topk (int): number of top predictions to return
Returns
"""
real_class = image_path.split(splitter)[-2]
# Convert to pytorch tensor
img_tensor = process_image(image_path)
if train_on_gpu:
img_tensor = img_tensor.view(1, 3, 224, 224).cuda()
else:
img_tensor = img_tensor.view(1, 3, 224, 224)
# Set to evaluation
with torch.no_grad():
model.eval()
# Model outputs log probabilities
out = model(img_tensor)
ps = torch.exp(out)
# Find the topk predictions
topk, topclass = ps.topk(topk, dim=1)
# Extract the actual classes and probabilities
top_classes = [
model.idx_to_class[class_] for class_ in topclass.cpu().numpy()[0]
]
top_p = topk.cpu().numpy()[0]
return img_tensor.cpu().squeeze(), top_p, top_classes, real_class
def accuracy(output, target, topk=[1]):
"""Compute the topk accuracy(s)"""
if train_on_gpu:
output = output.to('cuda')
target = target.to('cuda')
with torch.no_grad():
maxk = max(topk)
batch_size = target.size(0)
# Find the predicted classes and transpose
_, pred = output.topk(k=maxk, dim=1, largest=True, sorted=True)
pred = pred.t()
# Determine predictions equal to the targets
correct = pred.eq(target.view(1, -1).expand_as(pred))
res = []
# For each k, find the percentage of correct
for k in topk:
correct_k = correct[:k].view(-1).float().sum(0, keepdim=True)
res.append(correct_k.mul_(100.0 / batch_size).item())
return res