基于keras认识深度学习

发表于 | 分类于 | | 本文总阅读量 次

为什么选择

  • 编程友好：在学习深度学习原理过程中，最好自己代码实践，keras极度适合初学者，能快速实验

安装

ps:直接导入环境配置文件，命令conda env update -f=keras_environment.yml

image.png

定义模型

  • Sequential()模型

    1
    2
    3
    model = Sequential()
    model.add(Dense(32, input_shape=(784,)))
    model.add(Activation('relu'))

  • Functional()模型

    1
    2
    3
    a = Input(shape=(32,))
    b = Dense(32)(a)
    model = Model(inputs=a, outputs=b)

编译模型

1
model.compile(optimizer='sgd', loss='mse')

必须传入损失函数和优化算法

训练模型

1
history = model.fit(X, y, batch_size=10, epochs=100)

使用反向传播算法对数据进行训练

评估模型

1
loss, accuracy = model.evaluate(X, y)

模型预测

1
predictions = model.predict(x)

线性回归

Keras对波士顿房价进行回归

  1. 定义模型假设为y=Wx+b
  2. 编译模型时选择损失函数为均方误差函数，优化算法选择随机梯度下降
  3. 传入数据以及迭代此时
  4. 均方误差函数进行评估
  5. 进行预测
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
from keras.datasets import boston_housing
from keras.models import Sequential
from keras.layers import Dense
from keras.optimizers import SGD
import matplotlib.pyplot as plt
import numpy as np
# Boston房屋价格回归数据集
# 每条数据含有13个属性，目标值是该位置房子的房价中位数
(x_train, y_train), (x_test, y_test) = boston_housing.load_data()
# 构建模型
# Dense就是常用的全连接层，所实现的运算是output = activation(dot(input, kernel)+bias)
model = Sequential()
model.add(Dense(1,input_dim=13))
# 编译
model.compile(optimizer=SGD(lr=0.01,clipnorm=1),loss='mse',metrics=['mse'])
# 训练
cost = model.fit(x_train,y_train,epochs=20)
# 评估
score = model.evaluate(x_test,y_test)
#预测
y_pred = model.predict(x_test[0:1,:])
print(y_pred)
Epoch 1/20
404/404 [==============================] - 0s 527us/step - loss: 13144.4827 - mean_squared_error: 13144.4827
Epoch 2/20
404/404 [==============================] - 0s 40us/step - loss: 3041.2153 - mean_squared_error: 3041.2153
Epoch 3/20
404/404 [==============================] - 0s 45us/step - loss: 668.1933 - mean_squared_error: 668.1933
Epoch 4/20
404/404 [==============================] - 0s 42us/step - loss: 255.8141 - mean_squared_error: 255.8141
Epoch 5/20
404/404 [==============================] - 0s 40us/step - loss: 185.1011 - mean_squared_error: 185.1011
Epoch 6/20
404/404 [==============================] - 0s 40us/step - loss: 170.2183 - mean_squared_error: 170.2183
Epoch 7/20
404/404 [==============================] - 0s 40us/step - loss: 165.1749 - mean_squared_error: 165.1749
Epoch 8/20
404/404 [==============================] - 0s 37us/step - loss: 161.2603 - mean_squared_error: 161.2603
Epoch 9/20
404/404 [==============================] - 0s 35us/step - loss: 151.0976 - mean_squared_error: 151.0976
Epoch 10/20
404/404 [==============================] - 0s 35us/step - loss: 147.1785 - mean_squared_error: 147.1785
Epoch 11/20
404/404 [==============================] - 0s 40us/step - loss: 140.4976 - mean_squared_error: 140.4976
Epoch 12/20
404/404 [==============================] - 0s 42us/step - loss: 140.6603 - mean_squared_error: 140.6603
Epoch 13/20
404/404 [==============================] - 0s 40us/step - loss: 132.6787 - mean_squared_error: 132.6787
Epoch 14/20
404/404 [==============================] - 0s 40us/step - loss: 128.5385 - mean_squared_error: 128.5385
Epoch 15/20
404/404 [==============================] - 0s 37us/step - loss: 123.0882 - mean_squared_error: 123.0882
Epoch 16/20
404/404 [==============================] - 0s 40us/step - loss: 118.4096 - mean_squared_error: 118.4096
Epoch 17/20
404/404 [==============================] - 0s 40us/step - loss: 123.5432 - mean_squared_error: 123.5432
Epoch 18/20
404/404 [==============================] - 0s 42us/step - loss: 115.5308 - mean_squared_error: 115.5308
Epoch 19/20
404/404 [==============================] - 0s 40us/step - loss: 112.0784 - mean_squared_error: 112.0784
Epoch 20/20
404/404 [==============================] - 0s 37us/step - loss: 114.4626 - mean_squared_error: 114.4626
102/102 [==============================] - 0s 324us/step
[[15.296406]]

查看模型

1
2
3
4
from IPython.display import SVG
from keras.utils.vis_utils import model_to_dot

SVG(model_to_dot(model,show_shapes=True).create(prog='dot', format='svg'))

svg

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
### 单变量线性回归


# 构造数据
x=np.linspace(-10,10,100)
np.random.shuffle(x)
y=5*x+10+np.random.normal(0,5,(100,))#加入噪声
x_train,y_train,x_test,y_test = x[:80],y[:80],x[80:],y[80:]#训练数据与测试数据8：2
# 构建模型
# Dense就是常用的全连接层，所实现的运算是output = activation(dot(input, kernel)+bias)
model = Sequential()
model.add(Dense(1,input_dim=1))
# 编译
model.compile(optimizer='sgd',loss='mse',metrics=['mse'])
# 训练
cost = model.fit(x_train,y_train,epochs=20,verbose=0)
# 评估
score = model.evaluate(x_test,y_test)
#预测
y_pred = model.predict(x_test)
# 可视化
y_pred=model.predict(x_test)
plt.scatter(x_test,y_test,)
plt.plot(x_test,y_pred,'r')
plt.show()
20/20 [==============================] - 0s 1ms/step

png

1
2
3
4
5
6
7
8
9
10
11
12
# 正规方程求解(没有添加偏置)
def normal(x, y):
    #theta = np.linalg.inv(X.T@X)@X.T@y#X.T@X等价于X.T.dot(X)
    theta =np.linalg.inv(x.T.dot(x)).dot(x.T).dot(y)
    return theta

theta=normal(np.array(x_train).reshape(80,1),y_train)
# 可视化
plt.scatter(x_test,y_test)
y_pred= np.array(x_test)*theta
plt.plot(x_test,y_pred,'r')
plt.show()

png

image.png

逻辑回归

1
2
3
4
import numpy as np
from keras.models import Sequential
from keras.layers import Dense
import keras
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
#生成数据
x_train = np.random.random((1000, 20))
y_train = np.random.randint(2,size=(1000,1))
x_test = np.random.random((100, 20))
y_test = np.random.randint(2, size=(100, 1))

model = Sequential()
model.add(Dense(1,activation='sigmoid',input_dim=20))
model.compile(loss='binary_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])

cost = model.fit(x_train,y_train,epochs=20,callbacks=cbks,validation_data=(x_test, y_test),batch_size=128)
score = model.evaluate(x_test, y_test, batch_size=128)
score

image.png

神经网络

1
2
3
4
5
6
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation
from keras.optimizers import SGD
import tensorflow as tf
import keras
import numpy as np
C:\Users\DataMesh\Anaconda3\envs\py3\lib\site-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.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
x_train = np.random.random((1000, 20))
y_train = keras.utils.to_categorical(np.random.randint(10, size=(1000, 1)), num_classes=10)
x_test = np.random.random((100, 20))
y_test = keras.utils.to_categorical(np.random.randint(10, size=(100, 1)), num_classes=10)
print(x_train.shape,y_train.shape)
model = Sequential()

model.add(Dense(64, activation='relu', input_dim=20))
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))

sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='categorical_crossentropy',
              optimizer=sgd,
              metrics=['accuracy'])

model.fit(x_train, y_train,
          epochs=20,
          batch_size=128)
score = model.evaluate(x_test, y_test, batch_size=128)
model.summary()
(1000, 20) (1000, 10)
Epoch 1/20
1000/1000 [==============================] - 0s 313us/step - loss: 2.3688 - acc: 0.0980
Epoch 2/20
1000/1000 [==============================] - 0s 22us/step - loss: 2.3446 - acc: 0.1000
Epoch 3/20
1000/1000 [==============================] - 0s 20us/step - loss: 2.3213 - acc: 0.1250
Epoch 4/20
1000/1000 [==============================] - 0s 20us/step - loss: 2.3287 - acc: 0.1100
Epoch 5/20
1000/1000 [==============================] - 0s 21us/step - loss: 2.3065 - acc: 0.1140
Epoch 6/20
1000/1000 [==============================] - 0s 20us/step - loss: 2.3219 - acc: 0.1000
Epoch 7/20
1000/1000 [==============================] - 0s 20us/step - loss: 2.3094 - acc: 0.1180
Epoch 8/20
1000/1000 [==============================] - 0s 19us/step - loss: 2.3052 - acc: 0.1100
Epoch 9/20
1000/1000 [==============================] - 0s 19us/step - loss: 2.2990 - acc: 0.1370
Epoch 10/20
1000/1000 [==============================] - 0s 18us/step - loss: 2.2996 - acc: 0.1190
Epoch 11/20
1000/1000 [==============================] - 0s 19us/step - loss: 2.2975 - acc: 0.1320
Epoch 12/20
1000/1000 [==============================] - 0s 19us/step - loss: 2.3026 - acc: 0.1090
Epoch 13/20
1000/1000 [==============================] - 0s 19us/step - loss: 2.2979 - acc: 0.1130
Epoch 14/20
1000/1000 [==============================] - 0s 19us/step - loss: 2.2977 - acc: 0.1200
Epoch 15/20
1000/1000 [==============================] - 0s 20us/step - loss: 2.2874 - acc: 0.1100
Epoch 16/20
1000/1000 [==============================] - 0s 21us/step - loss: 2.2853 - acc: 0.1390
Epoch 17/20
1000/1000 [==============================] - 0s 19us/step - loss: 2.2897 - acc: 0.1330
Epoch 18/20
1000/1000 [==============================] - 0s 19us/step - loss: 2.2823 - acc: 0.1380
Epoch 19/20
1000/1000 [==============================] - 0s 18us/step - loss: 2.2890 - acc: 0.1290
Epoch 20/20
1000/1000 [==============================] - 0s 20us/step - loss: 2.2891 - acc: 0.1260
100/100 [==============================] - 0s 420us/step
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_3 (Dense)              (None, 64)                1344      
_________________________________________________________________
dropout_2 (Dropout)          (None, 64)                0         
_________________________________________________________________
dense_4 (Dense)              (None, 10)                650       
=================================================================
Total params: 1,994
Trainable params: 1,994
Non-trainable params: 0
_________________________________________________________________
  • 启动tensorboard，代码中加入
    1
    2
    3
    4
    5
    6
    7
    8
    9
    tb_cb=keras.callbacks.TensorBoard(log_dir='./logs', histogram_freq=1, write_graph=True, write_images=False, embeddings_freq=0, embeddings_layer_names=None, embeddings_metadata=None)
    cbks=[];
    cbks.append(tb_cb);
    model.fit(x_train, y_train,
              epochs=20,
              batch_size=128,
             verbose=0,
              validation_split=0.2,
             callbacks=cbks)

image.png