元框架(metaframework)。
TFLearn。模塊化深度學(xué)習(xí)框架，更高級(jí)API，快速實(shí)驗(yàn)，完全透明兼容。
TFLearn實(shí)現(xiàn)AlexNet。
https://github.com/tflearn/tflearn/blob/master/examples/images/alexnet.py
牛津大學(xué)鮮花數(shù)據(jù)集(Flower Dataset)。http://www.robots.ox.ac.uk/~vgg/data/flowers/17/ 。提供17個(gè)類別鮮花數(shù)據(jù)，每個(gè)類別80張圖片，有大量姿態(tài)、光變化。
# -*- coding: utf-8 -*-
""" AlexNet.
Applying 'Alexnet' to Oxford's 17 Category Flower Dataset classification task.
References:
- Alex Krizhevsky, Ilya Sutskever & Geoffrey E. Hinton. ImageNet
Classification with Deep Convolutional Neural Networks. NIPS, 2012.
- 17 Category Flower Dataset. Maria-Elena Nilsback and Andrew Zisserman.
Links:
- [AlexNet Paper](http://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf)
- [Flower Dataset (17)](http://www.robots.ox.ac.uk/~vgg/data/flowers/17/)
"""
from __future__ import division, print_function, absolute_import
import tflearn
from tflearn.layers.core import input_data, dropout, fully_connected
from tflearn.layers.conv import conv_2d, max_pool_2d
from tflearn.layers.normalization import local_response_normalization
from tflearn.layers.estimator import regression
import tflearn.datasets.oxflower17 as oxflower17
# 加載數(shù)據(jù)
X, Y = oxflower17.load_data(one_hot=True, resize_pics=(227, 227))
# Building 'AlexNet' 構(gòu)建網(wǎng)絡(luò)模型
network = input_data(shape=[None, 227, 227, 3])
network = conv_2d(network, 96, 11, strides=4, activation='relu')
network = max_pool_2d(network, 3, strides=2)
network = local_response_normalization(network)
network = conv_2d(network, 256, 5, activation='relu')
network = max_pool_2d(network, 3, strides=2)
network = local_response_normalization(network)
network = conv_2d(network, 384, 3, activation='relu')
network = conv_2d(network, 384, 3, activation='relu')
network = conv_2d(network, 256, 3, activation='relu')
network = max_pool_2d(network, 3, strides=2)
network = local_response_normalization(network)
network = fully_connected(network, 4096, activation='tanh')
network = dropout(network, 0.5)
network = fully_connected(network, 4096, activation='tanh')
network = dropout(network, 0.5)
network = fully_connected(network, 17, activation='softmax')
network = regression(network, optimizer='momentum',
loss='categorical_crossentropy',
learning_rate=0.001)
# Training 訓(xùn)練模型 載入AlexNet模型檢查點(diǎn)文件
model = tflearn.DNN(network, checkpoint_path='model_alexnet',
max_checkpoints=1, tensorboard_verbose=2)
model.fit(X, Y, n_epoch=1000, validation_set=0.1, shuffle=True,
show_metric=True, batch_size=64, snapshot_step=200,
snapshot_epoch=False, run_id='alexnet_oxflowers17')
Keras。高級(jí)Python神經(jīng)網(wǎng)絡(luò)框架。https://keras.io/。TensorFlow默認(rèn)框架。快速搭建原型。兼容Theano和TensorFlow。Keras高度封裝，適合新手，代碼更新快，示例代碼多，文檔、討論區(qū)完善。自動(dòng)調(diào)用GPU并行計(jì)算。模塊化，模型神經(jīng)層、成本函數(shù)、優(yōu)化器、初始化、激活函數(shù)、規(guī)范化模塊，組合創(chuàng)建模型。極簡(jiǎn)。易擴(kuò)展，容易添加新模塊。Python語(yǔ)言。
Keras模型。Keras核心數(shù)據(jù)結(jié)構(gòu)是模型。模型組織網(wǎng)絡(luò)層。Sequential模型，網(wǎng)絡(luò)層順序構(gòu)成棧，單輸入單輸出，層間只有相鄰關(guān)系，簡(jiǎn)單模型。Model模型建立復(fù)雜模型。
Sequential模型。加載完數(shù)據(jù)，X_train,Y_train,X_test,Y_test。構(gòu)建模型：
from keras.models import Sequential
from keras.layers.core import Dense, Dropout,Activation
model = Sequential()
model.add(Dense(output_dim=64, input_dim=100))
model.add(Activation('relu'))
model.add(Dense(output_dim=10))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='sgd', metrics='accuracy')
model.fit(X_train, Y_train, batch_size=32, nb_epoch=5)
loss_and_metrics = model.evaluate(X_test, Y_test, batch_size=32)
Keras源碼示例 https://github.com/fchollet/keras ：CIFAR10-圖片分類(CNN、實(shí)時(shí)數(shù)據(jù))、IMDB-電影評(píng)論觀點(diǎn)分類(LSTM)、Reuters-新聞主題分類(多層感知器)、MNIST-手寫數(shù)字識(shí)別(多層感知器、CNN)、OCR-識(shí)別字符級(jí)文本生成(LSTM)。
安裝。pip install keras 。選擇依賴后端，~/.keras/keras.json 修改最后一行"backend":"fensorflow" 。
Keras實(shí)現(xiàn)卷積神經(jīng)網(wǎng)絡(luò)。
https://github.com/fchollet/blob/master/examples/mnist_cnn.py 。
#!/usr/bin/python
# -*- coding:utf8 -*-
'''Trains a simple convnet on the MNIST dataset.
Gets to 99.25% test accuracy after 12 epochs
(there is still a lot of margin for parameter tuning).
16 seconds per epoch on a GRID K520 GPU.
'''
from __future__ import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K
batch_size = 128
num_classes = 10 # 分類數(shù)
epochs = 12 # 訓(xùn)練輪數(shù)
# input image dimensions 輸入圖片維度
img_rows, img_cols = 28, 28
# the data, shuffled and split between train and test sets
# 加載數(shù)據(jù)
(x_train, y_train), (x_test, y_test) = mnist.load_data()
if K.image_data_format() == 'channels_first':
#使用Theano順序:(conv_dim1,channels,conv_dim2,conv_dim3)
x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
input_shape = (1, img_rows, img_cols)
else:
#使用TensorFlow順序:(conv_dim1conv_dim2,conv_dim3,channels,)
x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# convert class vectors to binary class matrices 將類向量轉(zhuǎn)換為二進(jìn)制類矩陣
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
#模型構(gòu)建:2個(gè)卷積層、1個(gè)池化層、2個(gè)全連接層
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
activation='relu',
input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))
# 模型編譯
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adadelta(),
metrics=['accuracy'])
# 模型訓(xùn)練
model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_test, y_test))
# 模型評(píng)估
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
模型加載保存。 https://github.com/fchollet/keras/blob/master/tests/test_model_saving.py 。
Keras的save_model、load_model方法將模型、權(quán)重保存到HDF5文件。包括模型結(jié)構(gòu)、權(quán)重、訓(xùn)練配置(損失函數(shù)、優(yōu)化器)。方便中斷后再繼續(xù)訓(xùn)練。
from keras.models import save_model, load_model
def test_sequential_model_saving():
model = Sequential()
model.add(Dense(2, input_shape=(3,)))
model.add(RepeatVector(3))
model.add(TimeDistributed(Dense(3)))
model.compile(loss=losses.MSE,
optimizer=optimizers.RMSprop(lr=0.0001),
metrics=[metrics.categorical_accuracy],
sample_weight_mode='temporal')
x = np.random.random((1, 3))
y = np.random.random((1, 3, 3))
model.train_on_batch(x, y)
out = model.predict(x)
_, fname = tempfile.mkstemp('.h5') # 創(chuàng)建HDFS 5文件
save_model(model, fname)
new_model = load_model(fname)
os.remove(fname)
out2 = new_model.predict(x)
assert_allclose(out, out2, atol=1e-05)
# test that new updates are the same with both models
# 檢測(cè)新保存的模型和之前定義的模型是否一致
x = np.random.random((1, 3))
y = np.random.random((1, 3, 3))
model.train_on_batch(x, y)
new_model.train_on_batch(x, y)
out = model.predict(x)
out2 = new_model.predict(x)
assert_allclose(out, out2, atol=1e-05)
只保存模型結(jié)構(gòu)。
json_string = model.to_jsion()
yaml_string = model.to_yaml()
手動(dòng)編輯。
from keras.models import model_from_json
model = model_from_json(json_string)
model = model_from_yaml(yaml_string)
只保存模型權(quán)重。
model.save_weights('my_model_weights.h5')
model.load_weights('my_model_weights.h5')
參考資料：
《TensorFlow技術(shù)解析與實(shí)戰(zhàn)》
歡迎推薦上海機(jī)器學(xué)習(xí)工作機(jī)會(huì)，我的微信：qingxingfengzi
