-
Notifications
You must be signed in to change notification settings - Fork 0
/
deeplearning1.py
447 lines (367 loc) · 14.3 KB
/
deeplearning1.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
# -*- coding: utf-8 -*-
"""DeepLearning1.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1AbBtb4z3lE8IT3UR6DUX1LDK8Hv3VM2k
Instrucciones: descargar la siguiente base de datos de imágenes para clasificación de Kaggle
https://www.kaggle.com/ihelon/lego-minifigures-classification
"""
#Bibliotecas que vamos a necesitar
import os
import math
import time
import random
import numpy as np
import pandas as pd
import cv2
import matplotlib.pyplot as plt
import seaborn as sn
import albumentations as A
import torch
from torch.utils import data as torch_data
from torch import nn as torch_nn
import torchvision
from sklearn import metrics as sk_metrics
"""Vamos a utilizar la información dentro de nuestro Drive"""
#Monta los archivos de nuestro Drive para que puedan usarse en Colab
from google.colab import drive
drive.mount('/content/drive/')
# Establecemos el directorio de la base de datos
BASE_DIR = '/content/drive/My Drive/BasesDeDatos/lego/'
#Función para establecer la semilla de números aleatorios utilizando Pytorch
def set_seed(seed):
random.seed(seed)
os.environ["PYTHONHASHSEED"] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
#Esto es algo parecido a cuando configurabamos el random state en sklearn.
SEED = 42
set_seed(SEED)
# Leemos la información asociada a la base de datos de imágenes
df = pd.read_csv(BASE_DIR+'index.csv', index_col=0)
df.head()
# Obtener los objetos de entrenamiento
tmp_train = df[df['train-valid'] == 'train']
# Obtener las rutas de los objetos de entrenamiento
train_paths = tmp_train['path'].values
# Obtener las etiquetas asociadas a cada objeto (característica de salida)
train_targets = tmp_train['class_id'].values - 1
# Crear la ruta completa (directorio base + archivo en concreto)
train_paths = list(map(lambda x: os.path.join(BASE_DIR, x), train_paths))
print(train_paths)
# Obtener los objetos de prueba
tmp_valid = df[df['train-valid'] == 'valid']
# Obtener las rutas de los objetos de prueba
valid_paths = tmp_valid['path'].values
# Obtener las etiquetas asociadas a cada objeto (característica de salida)
valid_targets = tmp_valid['class_id'].values - 1
# Crear la ruta completa (directorio base + archivo en concreto)
valid_paths = list(map(lambda x: os.path.join(BASE_DIR, x), valid_paths))
print(valid_paths)
#Identificar el número de clases entrenamiento
n_classes = len(np.unique(valid_targets))
print('Número de clases: ', n_classes)
#Identificar el número de clases prueba
n_classes = len(np.unique(train_targets))
print('Número de clases: ', n_classes)
#Procesamiento inicial de las imágenes
class DataRetriever(torch_data.Dataset):
def __init__(
self,
paths,
targets,
image_size=(224, 224),
transforms=None
):
self.paths = paths
self.targets = targets
self.image_size = image_size
self.transforms = transforms
self.preprocess = torchvision.transforms.Compose([
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize(
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]
),
])
def __len__(self):
return len(self.targets)
def __getitem__(self, index):
img = cv2.imread(self.paths[index])
img = cv2.resize(img, self.image_size)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
if self.transforms:
img = self.transforms(image=img)['image']
img = self.preprocess(img)
y = torch.tensor(self.targets[index], dtype=torch.long)
return {'X': img, 'y': y}
#Aplicamos algunas transformaciones a las imágenes para que el modelo generado sea más robusto
def get_train_transforms():
return A.Compose(
[
A.Rotate(limit=30, border_mode=cv2.BORDER_REPLICATE, p=0.5),
A.Cutout(num_holes=8, max_h_size=20, max_w_size=20, p=0.25),
A.Cutout(num_holes=8, max_h_size=20, max_w_size=20, p=0.25),
A.HorizontalFlip(p=0.5),
A.RandomContrast(limit=(-0.3, 0.3), p=0.5),
A.RandomBrightness(limit=(-0.4, 0.4), p=0.5),
A.Blur(p=0.25),
],
p=1.0
)
#Especificamos el tamaño estandar de las imágenes
IMAGE_SIZE = (512, 512)
#Aplicamos las transformación que programos para hacer más robusto el modelo en ambos conjuntos tanto en el entrenamiento como en el de prueba
train_data_retriever = DataRetriever(
train_paths,
train_targets,
image_size=IMAGE_SIZE,
transforms=get_train_transforms()
)
valid_data_retriever = DataRetriever(
valid_paths,
valid_targets,
image_size=IMAGE_SIZE,
)
#Definimos el tamaño de la muestra que vamos a utilizar para entrenar y validar
TRAIN_BATCH_SIZE = 4
VALID_BATCH_SIZE = 1
train_loader = torch_data.DataLoader(
train_data_retriever,
batch_size=TRAIN_BATCH_SIZE,
shuffle=True,
)
valid_loader = torch_data.DataLoader(
valid_data_retriever,
batch_size=VALID_BATCH_SIZE,
shuffle=False,
)
#Desnormalizamos la imagen para poder visualizarla
def denormalize_image(image):
return image * [0.229, 0.224, 0.225] + [0.485, 0.456, 0.406]
#Visualización de algunos lotes de imágenes dentro del entrenamiento
plt.figure(figsize=(16, 16))
for i_batch, batch in enumerate(train_loader):
images, labels = batch['X'], batch['y']
for i in range(len(images)):
plt.subplot(4, 4, 4 * i_batch + i + 1)
plt.imshow(denormalize_image(images[i].permute(1, 2, 0).numpy()))
plt.title(labels[i].numpy())
plt.axis('off')
if i_batch >= 3:
break
#Visualización de algunos lotes de imágenes dentro del entrenamiento
plt.figure(figsize=(16, 16))
for i_batch, batch in enumerate(valid_loader):
images, labels = batch['X'], batch['y']
plt.subplot(4, 4, i_batch + 1)
plt.imshow(denormalize_image(images[0].permute(1, 2, 0).numpy()))
plt.title(labels[0].numpy())
plt.axis('off')
if i_batch >= 15:
break
#Configuramos la red neuronal profunda que vamos a utilizar
def get_net():
net = torch.hub.load('pytorch/vision:v0.6.0', 'mobilenet_v2', pretrained=True)
net.classifier = torch_nn.Linear(in_features=1280, out_features=n_classes, bias=True)
return net
#Definimos la función de optimización para entrenar el modelo
class LossMeter:
def __init__(self):
self.avg = 0
self.n = 0
def update(self, val):
self.n += 1
# incremental update
self.avg = val / self.n + (self.n - 1) / self.n * self.avg
# Definimos la función para evaluar el modelo en este caso precisión
class AccMeter:
def __init__(self):
self.avg = 0
self.n = 0
def update(self, y_true, y_pred):
y_true = y_true.cpu().numpy().astype(int)
y_pred = y_pred.cpu().numpy().argmax(axis=1).astype(int)
last_n = self.n
self.n += len(y_true)
true_count = np.sum(y_true == y_pred)
# incremental update
self.avg = true_count / self.n + last_n / self.n * self.avg
#Definimos la configuración completa para generar el modelo
class Trainer:
def __init__(
self,
model,
device,
optimizer,
criterion,
loss_meter,
score_meter
):
self.model = model
self.device = device
self.optimizer = optimizer
self.criterion = criterion
self.loss_meter = loss_meter
self.score_meter = score_meter
self.best_valid_score = 0
self.n_patience = 0
def fit(self, epochs, train_loader, valid_loader, save_path, patience):
for n_epoch in range(1, epochs + 1):
self.info_message("EPOCH: {}", n_epoch)
train_loss, train_score, train_time = self.train_epoch(train_loader)
valid_loss, valid_score, valid_time = self.valid_epoch(valid_loader)
m = '[Epoch {}: Train] loss: {:.5f}, score: {:.5f}, time: {} s'
self.info_message(
m, n_epoch, train_loss, train_score, train_time
)
m = '[Epoch {}: Valid] loss: {:.5f}, score: {:.5f}, time: {} s'
self.info_message(
m, n_epoch, valid_loss, valid_score, valid_time
)
if self.best_valid_score < valid_score:
m = 'The score improved from {:.5f} to {:.5f}. Save model to "{}"'
self.info_message(
m, self.best_valid_score, valid_score, save_path
)
self.save_model(n_epoch, save_path)
self.best_valid_score = valid_score
self.n_patience = 0
else:
self.n_patience += 1
if self.n_patience >= patience:
m = "\nValid score didn't improve last {} epochs."
self.info_message(m, patience)
break
def train_epoch(self, train_loader):
self.model.train()
t = time.time()
train_loss = self.loss_meter()
train_score = self.score_meter()
for step, batch in enumerate(train_loader, 1):
images = batch['X'].to(self.device)
targets = batch['y'].to(self.device)
self.optimizer.zero_grad()
outputs = self.model(images)
loss = self.criterion(outputs, targets)
loss.backward()
train_loss.update(loss.detach().item())
train_score.update(targets, outputs.detach())
self.optimizer.step()
_loss, _score = train_loss.avg, train_score.avg
_time = int(time.time() - t)
m = '[Train {}/{}] loss: {:.5f}, score: {:.5f}, time: {} s'
self.info_message(
m, step, len(train_loader), _loss, _score, _time, end='\r'
)
self.info_message('')
return train_loss.avg, train_score.avg, int(time.time() - t)
def valid_epoch(self, valid_loader):
self.model.eval()
t = time.time()
valid_loss = self.loss_meter()
valid_score = self.score_meter()
for step, batch in enumerate(valid_loader, 1):
with torch.no_grad():
images = batch['X'].to(self.device)
targets = batch['y'].to(self.device)
outputs = self.model(images)
loss = self.criterion(outputs, targets)
valid_loss.update(loss.detach().item())
valid_score.update(targets, outputs)
_loss, _score = valid_loss.avg, valid_score.avg
_time = int(time.time() - t)
m = '[Valid {}/{}] loss: {:.5f}, score: {:.5f}, time: {} s'
self.info_message(
m, step, len(valid_loader), _loss, _score, _time, end='\r'
)
self.info_message('')
return valid_loss.avg, valid_score.avg, int(time.time() - t)
def save_model(self, n_epoch, save_path):
torch.save(
{
'model_state_dict': self.model.state_dict(),
'optimizer_state_dict': self.optimizer.state_dict(),
'best_valid_score': self.best_valid_score,
'n_epoch': n_epoch,
},
save_path,
)
@staticmethod
def info_message(message, *args, end='\n'):
print(message.format(*args), end=end)
epochs = 50
patience = 3
model_save_path = 'model.torch'
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = get_net()
model.to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.0001)
criterion = torch_nn.CrossEntropyLoss()
trainer = Trainer(
model,
device,
optimizer,
criterion,
LossMeter,
AccMeter
)
trainer.fit(
epochs,
train_loader,
valid_loader,
model_save_path,
patience
)
# Load the best model
checkpoint = torch.load(model_save_path)
model.load_state_dict(checkpoint['model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
best_valid_score = checkpoint['best_valid_score']
n_epoch = checkpoint['n_epoch']
model.eval();
# Guardamos las predicciones del modelo y las etiquetas reales
y_pred = []
y_valid = []
for batch in valid_loader:
y_pred.extend(model(batch['X'].to(device)).argmax(axis=-1).cpu().numpy())
y_valid.extend(batch['y'])
# Calculate needed metrics
print(f'Precisión obtenida utilizando las imágenes de prueba:\t{sk_metrics.accuracy_score(y_valid, y_pred)}')
# Cargamos la infromación relaciona a las clases a etiquetas que puedan entender cualquier persona
df_metadata = pd.read_csv('/content/drive/My Drive/BasesDeDatos/lego/metadata.csv')
labels = df_metadata['minifigure_name'].tolist()
# Calculamos la matriz de confusión
confusion_matrix = sk_metrics.confusion_matrix(y_valid, y_pred)
df_confusion_matrix = pd.DataFrame(confusion_matrix, index=labels, columns=labels)
# mostramos los resultados de la matriz de confusión
plt.figure(figsize=(12, 12))
sn.heatmap(df_confusion_matrix, annot=True, cbar=False, cmap='Oranges', linewidths=1, linecolor='black')
plt.xlabel('Etiquetas predichas', fontsize=15)
plt.xticks(fontsize=12)
plt.ylabel('Etiquetas verdaderas', fontsize=15)
plt.yticks(fontsize=12);
# Interpretamos las salidas de nuestro modelo para encontrar los objetos que fueron mal clasificados
error_images = []
error_label = []
error_pred = []
error_prob = []
for batch in valid_loader:
_X_valid, _y_valid = batch['X'], batch['y']
pred = torch.softmax(model(_X_valid.to(device)), axis=-1).detach().cpu().numpy()
pred_class = pred.argmax(axis=-1)
if pred_class != _y_valid.cpu().numpy():
error_images.extend(_X_valid)
error_label.extend(_y_valid)
error_pred.extend(pred_class)
error_prob.extend(pred.max(axis=-1))
# Mostramos las imágenes que nuestro modelo clasifico de forma incorrecta
plt.figure(figsize=(16, 16))
for ind, image in enumerate(error_images):
plt.subplot(math.ceil(len(error_images) / int(len(error_images) ** 0.5)), int(len(error_images) ** 0.5), ind + 1)
plt.imshow(denormalize_image(image.permute(1, 2, 0).numpy()))
plt.title(f'Predicción: {labels[error_pred[ind]]} ({error_prob[ind]:.2f}) Real: {labels[error_label[ind]]}')
plt.axis('off')