当前位置:   article > 正文

快速入门深度学习五步法_深度学习初学者学习步驟

深度学习初学者学习步驟

作为一个程序员,我们可以像学习编程一样学习深度学习模型开发。我们以 Keras 为例来说明。

我们可以用 5 步 + 4 种基本元素 + 9 种基本层结构,这 5-4-9 模型来总结。

5步法: 

1. 构造网络模型 
2. 编译模型 
3. 训练模型 
4. 评估模型 
5. 使用模型进行预测

 

 

4种基本元素: 

1. 网络结构:由10种基本层结构和其他层结构组成 
2. 激活函数:如relu, softmax。口诀: 最后输出用softmax,其余基本都用relu 
3. 损失函数:categorical_crossentropy多分类对数损失,binary_crossentropy对数损失,mean_squared_error平均方差损失, mean_absolute_error平均绝对值损失 
4. 优化器:如SGD随机梯度下降, RMSProp, Adagrad, Adam, Adadelta等

 

 

9种基本层模型

 

包括3种主模型: 

1. 全连接层Dense 
2. 卷积层:如conv1d, conv2d 
3. 循环层:如lstm, gru

 

3种辅助层: 
1. Activation层 
2. Dropout层 
3. 池化层

 

3种异构网络互联层: 

1. 嵌入层:用于第一层,输入数据到其他网络的转换 
2. Flatten层:用于卷积层到全连接层之间的过渡 
3. Permute层:用于RNN与CNN之间的接口

我们通过一张图来理解下它们之间的关系

 

 

 

▌五步法

五步法是用深度学习来解决问题的五个步骤: 

1. 构造网络模型 
2. 编译模型 
3. 训练模型 
4. 评估模型 
5. 使用模型进行预测

在这五步之中,其实关键的步骤主要只有第一步,这一步确定了,后面的参数都可以根据它来设置。

过程化方法构造网络模型

我们先学习最容易理解的,过程化方法构造网络模型的过程。 

Keras中提供了Sequential容器来实现过程式构造。只要用Sequential的add方法把层结构加进来就可以了。10种基本层结构我们会在后面详细讲。

例:

 

from keras.models import Sequential
from keras.layers import Dense, Activation

model = Sequential()

model.add(Dense(units=64, input_dim=100))
model.add(Activation("relu"))
model.add(Dense(units=10))
model.add(Activation("softmax"))

 

对于什么样的问题构造什么样的层结构,我们会在后面的例子中介绍。

编译模型

模型构造好之后,下一步就可以调用Sequential的compile方法来编译它。

 

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

 

编译时需要指定两个基本元素:loss是损失函数,optimizer是优化函数。

如果只想用最基本的功能,只要指定字符串的名字就可以了。如果想配置更多的参数,调用相应的类来生成对象。例:我们想为随机梯度下降配上Nesterov动量,就生成一个SGD的对象就好了:

 

from keras.optimizers import SGD
model.compile(loss='categorical_crossentropy', optimizer=SGD(lr=0.01, momentum=0.9, nesterov=True))

lr是学习率,learning rate。

 

训练模型

调用fit函数,将输出的值X,打好标签的值y,epochs训练轮数,batch_size批次大小设置一下就可以了:

 

model.fit(x_train, y_train, epochs=5, batch_size=32)

 

评估模型

模型训练的好不好,训练数据不算数,需要用测试数据来评估一下:

 

loss_and_metrics = model.evaluate(x_test, y_test, batch_size=128)

 

用模型来预测

一切训练的目的是在于预测:

 

classes = model.predict(x_test, batch_size=128)

 

 

▌4种基本元素

网络结构

主要用后面的层结构来拼装。网络结构如何设计呢? 可以参考论文,比如这篇中不管是左边的19层的VGG-19,还是右边34层的resnet,只要按图去实现就好了。

激活函数

  • 对于多分类的情况,最后一层是softmax。 

  • 其它深度学习层中多用relu。 

  • 二分类可以用sigmoid。 

  • 另外浅层神经网络也可以用tanh。

损失函数

  • categorical_crossentropy:多分类对数损失

  • binary_crossentropy:对数损失

  • mean_squared_error:均方差

  • mean_absolute_error:平均绝对值损失

对于多分类来说,主要用categorical_crossentropy。

 

优化器

  • SGD:随机梯度下降

  • Adagrad:Adaptive Gradient自适应梯度下降

  • Adadelta:对于Adagrad的进一步改进

  • RMSProp

  • Adam

本文将着重介绍后两种教程。

深度学习中的函数式编程

前面介绍的各种基本层,除了可以add进Sequential容器串联之外,它们本身也是callable对象,被调用之后,返回的还是callable对象。所以可以将它们视为函数,通过调用的方式来进行串联。

来个官方例子:

 

from keras.layers import Input, Dense
from keras.models import Model

inputs = Input(shape=(784,))

x = Dense(64, activation='relu')(inputs)
x = Dense(64, activation='relu')(x)
predictions = Dense(10, activation='softmax')(x)

model = Model(inputs=inputs, outputs=predictions)
model.compile(optimizer='rmsprop',
             loss='categorical_crossentropy',
             metrics=['accuracy'])
model.fit(data, labels)

为什么要用函数式编程?

答案是,复杂的网络结构并不是都是线性的add进容器中的。并行的,重用的,什么情况都有。这时候callable的优势就发挥出来了。

比如下面的Google Inception模型,就是带并联的:

 

我们的代码自然是以并联应对并联了,一个输入input_img被三个模型所重用:

 

from keras.layers import Conv2D, MaxPooling2D, Input

input_img = Input(shape=(256, 256, 3))

tower_1 = Conv2D(64, (1, 1), padding='same', activation='relu')(input_img)
tower_1 = Conv2D(64, (3, 3), padding='same', activation='relu')(tower_1)

tower_2 = Conv2D(64, (1, 1), padding='same', activation='relu')(input_img)
tower_2 = Conv2D(64, (5, 5), padding='same', activation='relu')(tower_2)

tower_3 = MaxPooling2D((3, 3), strides=(1, 1), padding='same')(input_img)
tower_3 = Conv2D(64, (1, 1), padding='same', activation='relu')(tower_3)

output = keras.layers.concatenate([tower_1, tower_2, tower_3], axis=1)

 

 

▌案例教程

CNN处理MNIST手写识别

光说不练是假把式。我们来看看符合五步法的处理MNIST的例子。

首先解析一下核心模型代码,因为模型是线性的,我们还是用Sequential容器

 

model = Sequential()

核心是两个卷积层:

 

model.add(Conv2D(32, kernel_size=(3, 3),
                activation='relu',
                input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))

为了防止过拟合,我们加上一个最大池化层,再加上一个Dropout层:

 

model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

下面要进入全连接层输出了,这两个中间的数据转换需要一个Flatten层:

 

model.add(Flatten())

下面是全连接层,激活函数是relu。 

还怕过拟合,再来个Dropout层!

 

model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))

最后通过一个softmax激活函数的全连接网络输出:

 

model.add(Dense(num_classes, activation='softmax'))

下面是编译这个模型,损失函数是categorical_crossentropy多类对数损失函数,优化器选用Adadelta。

 

model.compile(loss=keras.losses.categorical_crossentropy,
             optimizer=keras.optimizers.Adadelta(),
             metrics=['accuracy'])

下面是可以运行的完整代码:

 

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
epochs = 12


# input image dimensions
img_rows, img_cols = 28, 28


# the data, split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()

if K.image_data_format() == 'channels_first':
   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:
   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
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

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'))

model.compile(loss=keras.losses.categorical_crossentropy,
             optimizer=keras.optimizers.Adadelta(),
             metrics=['accuracy'])
model.fit(x_train, y_train,
         batch_size=batch_size,
         epochs=epochs,
         verbose=1,
         validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])

下面我们来个surprise例子,处理一下各种语言之间的翻译。

机器翻译:多语种互译

英译汉,汉译英之类的事情,在学生时代是不是一直难为这你呢? 

现在不用担心了,只要有两种语言的对照表,我们就可以训练一个模型来像做一个机器翻译。 

首先得下载一个字典:http://www.manythings.org/anki/

然后我们还是老办法,我们先看一下核心代码。没啥说的,这类序列化处理的问题用的一定是RNN,通常都是用LSTM. 

 

encoder_inputs = Input(shape=(None, num_encoder_tokens))
encoder = LSTM(latent_dim, return_state=True)
encoder_outputs, state_h, state_c = encoder(encoder_inputs)
encoder_states = [state_h, state_c]
decoder_inputs = Input(shape=(None, num_decoder_tokens))
decoder_lstm = LSTM(latent_dim, return_sequences=True, return_state=True)
decoder_outputs, _, _ = decoder_lstm(decoder_inputs,
                                    initial_state=encoder_states)
decoder_dense = Dense(num_decoder_tokens, activation='softmax')
decoder_outputs = decoder_dense(decoder_outputs)
model = Model([encoder_inputs, decoder_inputs], decoder_outputs)

 

优化器选用rmsprop,损失函数还是categorical_crossentropy. 

validation_split是将一个集合随机分成训练集和测试集。

 

# Run training
model.compile(optimizer='rmsprop', loss='categorical_crossentropy')
model.fit([encoder_input_data, decoder_input_data], decoder_target_data,
         batch_size=batch_size,
         epochs=epochs,
         validation_split=0.2)

 

最后,训练一个模型不容易,我们将其存储起来。

 

model.save('s2s.h5')

 

最后,附上完整的实现了机器翻译功能的代码,加上注释和空行有100多行,供有需要的同学取用。

 

from __future__ import print_function

from keras.models import Model
from keras.layers import Input, LSTM, Dense
import numpy as np

batch_size = 64  # Batch size for training.
epochs = 100  # Number of epochs to train for.
latent_dim = 256  # Latent dimensionality of the encoding space.
num_samples = 10000  # Number of samples to train on.
# Path to the data txt file on disk.
data_path = 'fra-eng/fra.txt'

# Vectorize the data.
input_texts = []
target_texts = []
input_characters = set()
target_characters = set()
with open(data_path, 'r', encoding='utf-8') as f:
   lines = f.read().split('\n')
for line in lines[: min(num_samples, len(lines) - 1)]:
   input_text, target_text = line.split('\t')
   # We use "tab" as the "start sequence" character
   # for the targets, and "\n" as "end sequence" character.
   target_text = '\t' + target_text + '\n'
   input_texts.append(input_text)
   target_texts.append(target_text)
   for char in input_text:
       if char not in input_characters:
           input_characters.add(char)
   for char in target_text:
       if char not in target_characters:
           target_characters.add(char)

input_characters = sorted(list(input_characters))
target_characters = sorted(list(target_characters))
num_encoder_tokens = len(input_characters)
num_decoder_tokens = len(target_characters)
max_encoder_seq_length = max([len(txt) for txt in input_texts])
max_decoder_seq_length = max([len(txt) for txt in target_texts])

print('Number of samples:', len(input_texts))
print('Number of unique input tokens:', num_encoder_tokens)
print('Number of unique output tokens:', num_decoder_tokens)
print('Max sequence length for inputs:', max_encoder_seq_length)
print('Max sequence length for outputs:', max_decoder_seq_length)

input_token_index = dict(
   [(char, i) for i, char in enumerate(input_characters)])
target_token_index = dict(
   [(char, i) for i, char in enumerate(target_characters)])

encoder_input_data = np.zeros(
   (len(input_texts), max_encoder_seq_length, num_encoder_tokens),
   dtype='float32')
decoder_input_data = np.zeros(
   (len(input_texts), max_decoder_seq_length, num_decoder_tokens),
   dtype='float32')
decoder_target_data = np.zeros(
   (len(input_texts), max_decoder_seq_length, num_decoder_tokens),
   dtype='float32')

for i, (input_text, target_text) in enumerate(zip(input_texts, target_texts)):
   for t, char in enumerate(input_text):
       encoder_input_data[i, t, input_token_index[char]] = 1.
   for t, char in enumerate(target_text):
       # decoder_target_data is ahead of decoder_input_data by one timestep
       decoder_input_data[i, t, target_token_index[char]] = 1.
       if t > 0:
           # decoder_target_data will be ahead by one timestep
           # and will not include the start character.
           decoder_target_data[i, t - 1, target_token_index[char]] = 1.

# Define an input sequence and process it.
encoder_inputs = Input(shape=(None, num_encoder_tokens))
encoder = LSTM(latent_dim, return_state=True)
encoder_outputs, state_h, state_c = encoder(encoder_inputs)
# We discard `encoder_outputs` and only keep the states.
encoder_states = [state_h, state_c]

# Set up the decoder, using `encoder_states` as initial state.
decoder_inputs = Input(shape=(None, num_decoder_tokens))
# We set up our decoder to return full output sequences,
# and to return internal states as well. We don't use the
# return states in the training model, but we will use them in inference.
decoder_lstm = LSTM(latent_dim, return_sequences=True, return_state=True)
decoder_outputs, _, _ = decoder_lstm(decoder_inputs,
                                    initial_state=encoder_states)
decoder_dense = Dense(num_decoder_tokens, activation='softmax')
decoder_outputs = decoder_dense(decoder_outputs)

# Define the model that will turn
# `encoder_input_data` & `decoder_input_data` into `decoder_target_data`
model = Model([encoder_inputs, decoder_inputs], decoder_outputs)

# Run training
model.compile(optimizer='rmsprop', loss='categorical_crossentropy')
model.fit([encoder_input_data, decoder_input_data], decoder_target_data,
         batch_size=batch_size,
         epochs=epochs,
         validation_split=0.2)
# Save model
model.save('s2s.h5')

encoder_model = Model(encoder_inputs, encoder_states)

decoder_state_input_h = Input(shape=(latent_dim,))
decoder_state_input_c = Input(shape=(latent_dim,))
decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c]
decoder_outputs, state_h, state_c = decoder_lstm(
   decoder_inputs, initial_state=decoder_states_inputs)
decoder_states = [state_h, state_c]
decoder_outputs = decoder_dense(decoder_outputs)
decoder_model = Model(
   [decoder_inputs] + decoder_states_inputs,
   [decoder_outputs] + decoder_states)

# Reverse-lookup token index to decode sequences back to
# something readable.
reverse_input_char_index = dict(
   (i, char) for char, i in input_token_index.items())
reverse_target_char_index = dict(
   (i, char) for char, i in target_token_index.items())


def decode_sequence(input_seq):
   # Encode the input as state vectors.
   states_value = encoder_model.predict(input_seq)

   # Generate empty target sequence of length 1.
   target_seq = np.zeros((1, 1, num_decoder_tokens))
   # Populate the first character of target sequence with the start character.
   target_seq[0, 0, target_token_index['\t']] = 1.

   # Sampling loop for a batch of sequences
   # (to simplify, here we assume a batch of size 1).
   stop_condition = False
   decoded_sentence = ''
   while not stop_condition:
       output_tokens, h, c = decoder_model.predict(
           [target_seq] + states_value)

       # Sample a token
       sampled_token_index = np.argmax(output_tokens[0, -1, :])
       sampled_char = reverse_target_char_index[sampled_token_index]
       decoded_sentence += sampled_char

       # Exit condition: either hit max length
       # or find stop character.
       if (sampled_char == '\n' or
          len(decoded_sentence) > max_decoder_seq_length):
           stop_condition = True

       # Update the target sequence (of length 1).
       target_seq = np.zeros((1, 1, num_decoder_tokens))
       target_seq[0, 0, sampled_token_index] = 1.

       # Update states
       states_value = [h, c]

   return decoded_sentence


for seq_index in range(100):
   # Take one sequence (part of the training set)
   # for trying out decoding.
   input_seq = encoder_input_data[seq_index: seq_index + 1]
   decoded_sentence = decode_sequence(input_seq)
   print('-')
   print('Input sentence:', input_texts[seq_index])
   print('Decoded sentence:', decoded_sentence)

 
声明:本文内容由网友自发贡献,不代表【wpsshop博客】立场,版权归原作者所有,本站不承担相应法律责任。如您发现有侵权的内容,请联系我们。转载请注明出处:https://www.wpsshop.cn/w/小蓝xlanll/article/detail/379818
推荐阅读
相关标签
  

闽ICP备14008679号