keras 入门建模型案例

原链接

In [1]:

import os
import cv2
import random
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

from tqdm import tqdm
from sklearn.utils import shuffle
from sklearn.model_selection import train_test_split

%matplotlib inline
%config InlineBackend.figure_format = 'retina'

导入数据

In [2]:

df_train = pd.read_csv('train.csv', header=None)
df_train.columns = ['image_id', 'class', 'label']
df_train.head()

Out[2]:

	image_id	class	label
0	Images/collar_design_labels/4d8a38b29930a403e5...	collar_design_labels	nnynn
1	Images/collar_design_labels/bd0981f231180d2b00...	collar_design_labels	nynnn
2	Images/collar_design_labels/26937e1724feadfe39...	collar_design_labels	ynnnn
3	Images/collar_design_labels/cf4140ec542887270f...	collar_design_labels	nynnn
4	Images/collar_design_labels/50644b2b9de045f2d1...	collar_design_labels	nynnn

In [3]:

classes = ['collar_design_labels', 'neckline_design_labels', 'skirt_length_labels', 
           'sleeve_length_labels', 'neck_design_labels', 'coat_length_labels', 'lapel_design_labels', 
           'pant_length_labels']
len(classes)

Out[3]:

8

加载数据

In [4]:

cur_class = classes[0]
df_load = df_train[(df_train['class'] == cur_class)].copy()
df_load.reset_index(inplace=True)
del df_load['index']

print('{0}: {1}'.format(cur_class, len(df_load)))
df_load.head()

collar_design_labels: 8393

Out[4]:

	image_id	class	label
0	Images/collar_design_labels/4d8a38b29930a403e5...	collar_design_labels	nnynn
1	Images/collar_design_labels/bd0981f231180d2b00...	collar_design_labels	nynnn
2	Images/collar_design_labels/26937e1724feadfe39...	collar_design_labels	ynnnn
3	Images/collar_design_labels/cf4140ec542887270f...	collar_design_labels	nynnn
4	Images/collar_design_labels/50644b2b9de045f2d1...	collar_design_labels	nynnn

In [5]:

df_load[(df_load.index == 2)]

Out[5]:

	image_id	class	label
2	Images/collar_design_labels/26937e1724feadfe39...	collar_design_labels	ynnnn

In [6]:

n = len(df_load)
n_class = len(df_load['label'][0])
width = 299

X = np.zeros((n, width, width, 3), dtype=np.uint8)
y = np.zeros((n, n_class), dtype=np.uint8)

for i in tqdm(range(n)):
    tmp_label = df_load['label'][i]
    if len(tmp_label) > n_class:
        print(df_load['image_id'][i])
    X[i] = cv2.resize(cv2.imread('./train/{0}'.format(df_load['image_id'][i])), (width, width))
    y[i][tmp_label.find('y')] = 1

100%|██████████| 8393/8393 [00:31<00:00, 269.94it/s]

In [7]:

plt.figure(figsize=(12, 7))
for i in range(8):
    random_index = random.randint(0, n-1)
    plt.subplot(2, 4, i+1)
    plt.imshow(X[random_index][:,:,::-1])
    plt.title(y[random_index])

提取特征

In [8]:

from keras.layers import *
from keras.models import *
from keras.callbacks import *
from keras.optimizers import *
from keras.applications import *
from keras.regularizers import *
from keras.applications.inception_v3 import preprocess_input

/usr/local/lib/python3.5/dist-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Using TensorFlow backend.

In [9]:

cnn_model = InceptionResNetV2(include_top=False, input_shape=(width, width, 3), weights='imagenet')
inputs = Input((width, width, 3))

x = inputs
x = Lambda(preprocess_input, name='preprocessing')(x)
x = cnn_model(x)
x = GlobalAveragePooling2D()(x)
x = Dropout(0.5)(x)
x = Dense(n_class, activation='softmax', name='softmax')(x)

model = Model(inputs, x)

划分训练、测试集

In [10]:

X_train, X_valid, y_train, y_valid = train_test_split(X, y, test_size=0.12, random_state=42)
X_train.shape, y_train.shape

Out[10]:

((7385, 299, 299, 3), (7385, 5))

In [11]:

adam = Adam(lr=0.001)
prefix_cls = cur_class.split('_')[0]

model.compile(optimizer=adam,
              loss='categorical_crossentropy',
              metrics=['accuracy'])

checkpointer = ModelCheckpoint(filepath='./models/{0}.best.h5'.format(prefix_cls), verbose=1, 
                               save_best_only=True)

h = model.fit(X_train, y_train, batch_size=32, epochs=80, 
              callbacks=[EarlyStopping(patience=3), checkpointer], 
              shuffle=True, 
              validation_split=0.1)

Train on 6646 samples, validate on 739 samples
Epoch 1/80
6624/6646 [============================>.] - ETA: 0s - loss: 0.9380 - acc: 0.6532Epoch 00001: val_loss improved from inf to 1.10376, saving model to ./models/collar.best.h5
6646/6646 [==============================] - 196s 29ms/step - loss: 0.9367 - acc: 0.6539 - val_loss: 1.1038 - val_acc: 0.6766
Epoch 2/80
6624/6646 [============================>.] - ETA: 0s - loss: 0.6084 - acc: 0.7876Epoch 00002: val_loss improved from 1.10376 to 0.75603, saving model to ./models/collar.best.h5
6646/6646 [==============================] - 183s 27ms/step - loss: 0.6077 - acc: 0.7880 - val_loss: 0.7560 - val_acc: 0.7361
Epoch 3/80
6624/6646 [============================>.] - ETA: 0s - loss: 0.4251 - acc: 0.8552Epoch 00003: val_loss did not improve
6646/6646 [==============================] - 181s 27ms/step - loss: 0.4264 - acc: 0.8554 - val_loss: 1.5847 - val_acc: 0.5372
Epoch 4/80
6624/6646 [============================>.] - ETA: 0s - loss: 0.3442 - acc: 0.8819Epoch 00004: val_loss did not improve
6646/6646 [==============================] - 181s 27ms/step - loss: 0.3444 - acc: 0.8816 - val_loss: 0.8974 - val_acc: 0.7294
Epoch 5/80
6624/6646 [============================>.] - ETA: 0s - loss: 0.2424 - acc: 0.9179Epoch 00005: val_loss did not improve
6646/6646 [==============================] - 181s 27ms/step - loss: 0.2420 - acc: 0.9180 - val_loss: 1.4803 - val_acc: 0.6455

In [12]:

plt.figure(figsize=(10, 4))
plt.subplot(1, 2, 1)
plt.plot(h.history['loss'])
plt.plot(h.history['val_loss'])
plt.legend(['loss', 'val_loss'])
plt.ylabel('loss')
plt.xlabel('epoch')

plt.subplot(1, 2, 2)
plt.plot(h.history['acc'])
plt.plot(h.history['val_acc'])
plt.legend(['acc', 'val_acc'])
plt.ylabel('acc')
plt.xlabel('epoch')

Out[12]:

Text(0.5,0,'epoch')

In [14]:

model.evaluate(X_train, y_train, batch_size=256)

7385/7385 [==============================] - 62s 8ms/step

Out[14]:

[0.7887161796288307, 0.7599187538037853]

In [15]:

model.evaluate(X_valid, y_valid, batch_size=256)

1008/1008 [==============================] - 9s 9ms/step

Out[15]:

[1.4538595203369382, 0.6448412698412699]

加载测试集

In [16]:

df_test = pd.read_csv('test_a.csv', header=None)
df_test.columns = ['image_id', 'class', 'x']
del df_test['x']
df_test.head()

Out[16]:

	image_id	class
0	Images/collar_design_labels/60951f5761ea3a01d6...	collar_design_labels
1	Images/collar_design_labels/677e1183282769a3fe...	collar_design_labels
2	Images/collar_design_labels/3f5a2be205e5d253eb...	collar_design_labels
3	Images/collar_design_labels/1712bf3d7b57d779ad...	collar_design_labels
4	Images/collar_design_labels/0b2b4254f35ce3a41a...	collar_design_labels

In [17]:

df_load = df_test[(df_test['class'] == cur_class)].copy()
df_load.reset_index(inplace=True)
del df_load['index']

print('{0}: {1}'.format(cur_class, len(df_load)))
df_load.head()

collar_design_labels: 1082

Out[17]:

	image_id	class
0	Images/collar_design_labels/60951f5761ea3a01d6...	collar_design_labels
1	Images/collar_design_labels/677e1183282769a3fe...	collar_design_labels
2	Images/collar_design_labels/3f5a2be205e5d253eb...	collar_design_labels
3	Images/collar_design_labels/1712bf3d7b57d779ad...	collar_design_labels
4	Images/collar_design_labels/0b2b4254f35ce3a41a...	collar_design_labels

In [18]:

n = len(df_load)
X_test = np.zeros((n, width, width, 3), dtype=np.uint8)

for i in tqdm(range(n)):
    X_test[i] = cv2.resize(cv2.imread('./test_a/{0}'.format(df_load['image_id'][i])), (width, width))

100%|██████████| 1082/1082 [00:04<00:00, 265.47it/s]

In [19]:

test_np = model.predict(X_test, batch_size=256)

In [20]:

test_np.shape

Out[20]:

(1082, 5)

In [21]:

result = []

for i, row in df_load.iterrows():
    tmp_list = test_np[i]
    tmp_result = ''
    for tmp_ret in tmp_list:
        tmp_result += '{:.4f};'.format(tmp_ret)
        
    result.append(tmp_result[:-1])

df_load['result'] = result
df_load.head()

Out[21]:

	image_id	class	result
0	Images/collar_design_labels/60951f5761ea3a01d6...	collar_design_labels	0.0008;0.9845;0.0005;0.0141;0.0001
1	Images/collar_design_labels/677e1183282769a3fe...	collar_design_labels	0.0000;0.0000;0.0000;1.0000;0.0000
2	Images/collar_design_labels/3f5a2be205e5d253eb...	collar_design_labels	0.0000;0.9998;0.0000;0.0002;0.0000
3	Images/collar_design_labels/1712bf3d7b57d779ad...	collar_design_labels	0.0056;0.0004;0.0087;0.9846;0.0007
4	Images/collar_design_labels/0b2b4254f35ce3a41a...	collar_design_labels	0.0000;0.0000;0.0002;0.9998;0.0000

In [22]:

df_load.to_csv('./result/{}_0307a.csv'.format(prefix_cls), header=None, index=False)
prefix_cls

Out[22]:

'collar'