tensorflow2.0学习笔记：神经网络

mnist数据集分类：深度神经网络在网络中，激活函数、批归一化、Dropout的添加和使用。

import matplotlib as mpl
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import sklearn
import pandas as pd
import os
import tensorflow as tf

from tensorflow import keras

print(tf.__version__)

2.0.0

fashion_mnist = keras.datasets.fashion_mnist
(x_train_all,y_train_all),(x_test,y_test) = fashion_mnist.load_data()
x_valid,x_train = x_train_all[:5000],x_train_all[5000:]
y_valid,y_train = y_train_all[:5000],y_train_all[5000:]

print(x_valid.shape,y_valid.shape)
print(x_train.shape,y_train.shape)
print(x_test.shape,y_test.shape)

(5000, 28, 28) (5000,)
(55000, 28, 28) (55000,)
(10000, 28, 28) (10000,)

#归一化：x = (x-mu)/std 均值为0，方差为1
from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(
    x_train.astype(np.float32).reshape(-1,1)).reshape(-1,28,28)
x_valid_scaled = scaler.transform(
    x_valid.astype(np.float32).reshape(-1,1)).reshape(-1,28,28)
x_test_scaled = scaler.transform(
    x_test.astype(np.float32).reshape(-1,1)).reshape(-1,28,28)

print(np.max(x_train_scaled),np.min(x_train_scaled))

2.0231433 -0.8105136

深度神经网络太深：

1. 参数太多，训练不充分
2. 梯度下降过程中，梯度消失(复合函数，链式法则求导)
3. 批归一化缓减梯度消失

model = keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28,28])) #输入28*28的图像，展平成28*28的1维向量
#深度的神经网络20层
for _ in range(20):
    #激活函数selu自带归一化
    model.add(keras.layers.Dense(100,activation="selu"))
    #批归一化，每层都做
    model.add(keras.layers.BatchNormalization()) 
    '''
    #批归一化也可以放在激活函数前
    model.add(keras.layers.Dense(100))
    model.add(keras.layers.BatchNormalization())
    model.add(keras.layers.Activation(relu))
    '''

#Dropout:随机地禁用一些神经单元，作用：防止过拟合。
#modeld.add(keras.layers.Dropout(rate=0.5))

#AlphaDropout:1.均值和方差不变  2.归一化的性质不变    
model.add(keras.layers.AlphaDropout(rate=0.5))
model.add(keras.layers.Dense(10,activation="softmax"))

model.compile(loss="sparse_categorical_crossentropy",
             optimizer = "adam",
             metrics = ["accuracy"])

#callbacks: Tensorboard , earlystopping , ModelCheckpoint
#设置一个文件夹
logdir = os.path.join('./dnn_callbacks')
if not os.path.exists(logdir):
    os.mkdir(logdir)
output_model_file = os.path.join(logdir,"fashion_mnist_model.h5")

callbacks = [
#     keras.callbacks.TensorBoard(logdir),
    keras.callbacks.ModelCheckpoint(output_model_file,save_best_only = True),
    keras.callbacks.EarlyStopping(monitor="val_loss",patience=5,min_delta=1e-3)
]

history = model.fit(x_train_scaled,y_train,epochs=10,
                    validation_data=(x_valid_scaled,y_valid))

Train on 55000 samples, validate on 5000 samples
Epoch 1/10
55000/55000 [==============================] - 23s 424us/sample - loss: 0.4710 - accuracy: 0.8436 - val_loss: 0.6472 - val_accuracy: 0.8574
Epoch 2/10
55000/55000 [==============================] - 23s 421us/sample - loss: 0.4525 - accuracy: 0.8482 - val_loss: 0.5506 - val_accuracy: 0.8576
Epoch 3/10
55000/55000 [==============================] - 24s 434us/sample - loss: 0.4371 - accuracy: 0.8544 - val_loss: 0.5437 - val_accuracy: 0.8822
Epoch 4/10
55000/55000 [==============================] - 22s 403us/sample - loss: 0.4145 - accuracy: 0.8591 - val_loss: 0.6947 - val_accuracy: 0.8454
Epoch 5/10
55000/55000 [==============================] - 23s 421us/sample - loss: 0.4122 - accuracy: 0.8627 - val_loss: 0.5254 - val_accuracy: 0.8842
Epoch 6/10
55000/55000 [==============================] - 24s 445us/sample - loss: 0.3988 - accuracy: 0.8656 - val_loss: 0.5548 - val_accuracy: 0.8626
Epoch 7/10
55000/55000 [==============================] - 21s 386us/sample - loss: 0.3883 - accuracy: 0.8696 - val_loss: 0.5432 - val_accuracy: 0.8808
Epoch 8/10
55000/55000 [==============================] - 22s 406us/sample - loss: 0.3779 - accuracy: 0.8728 - val_loss: 0.5272 - val_accuracy: 0.8884
Epoch 9/10
55000/55000 [==============================] - 21s 388us/sample - loss: 0.3686 - accuracy: 0.8743 - val_loss: 0.5292 - val_accuracy: 0.8832
Epoch 10/10
55000/55000 [==============================] - 23s 421us/sample - loss: 0.3599 - accuracy: 0.8781 - val_loss: 0.5088 - val_accuracy: 0.8792

def plot_learning_curve(history):
    pd.DataFrame(history.history).plot(figsize=(8,5))
    plt.grid(True)
    plt.gca().set_ylim(0,1)
    plt.show()
#DateFrame 数据类型的作图
plot_learning_curve(history)

y = model.evaluate(x_test_scaled,y_test)
# - 1s 110us/sample - loss: 0.3048 - accuracy: 0.8724