fashion_mnist_用深度可分离卷积缩小模型,用sklearn的StandardScaler处理图片

import matplotlib.pyplot as plt
from sklearn.datasets import load_iris
from sklearn.decomposition import PCA
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler 
import tensorflow as tf
import numpy as np
import pandas as pd
import os

(X_train,y_train),(X_test,y_test)=tf.keras.datasets.fashion_mnist.load_data()

X_train.shape,X_test.shape,y_train.shape,y_test.shape

((60000, 28, 28), (10000, 28, 28), (60000,), (10000,))

将60000×28×28的数据变成二维数,组然后用sklearn的StandardScaler进行数据预处理 ,sklearn不能直接处理图片张量

X_train_1 =  StandardScaler().fit_transform(X_train.astype(np.float32).reshape(-1,1))

X_train_1.shape

(47040000, 1)

X_train_2 = X_train_1.reshape(-1,28,28)
X_train_2.shape

(60000, 28, 28)

X_train_2.shape[1:]

(28, 28)

X_test_2 =  StandardScaler().fit_transform(X_test.astype(np.float32).reshape(-1,1)).reshape(-1,28,28)
X_test_2.shape

(10000, 28, 28)

X_train = np.expand_dims(X_train_2,-1)
X_test = np.expand_dims(X_test_2,-1)

X_train.shape,X_test.shape

((60000, 28, 28, 1), (10000, 28, 28, 1))

y_train = y_train.astype(np.int32)
y_test = y_test.astype(np.int32)

y_train.shape,y_test.shape

((60000,), (10000,))

深度可分离卷积网络,自编码器结构

model = tf.keras.Sequential()
model.add(tf.keras.layers.Conv2D(filters=64,kernel_size=(3,3),
                                 padding='same',activation='selu',
                                 input_shape=(X_train.shape[1:])))
model.add(tf.keras.layers.MaxPool2D())
model.add(tf.keras.layers.SeparableConv2D(filters=32,kernel_size=(3,3),padding='same',activation='selu'))
model.add(tf.keras.layers.MaxPool2D())
model.add(tf.keras.layers.SeparableConv2D(filters=16,kernel_size=(3,3),padding='same',activation='selu'))
model.add(tf.keras.layers.MaxPool2D())
model.add(tf.keras.layers.SeparableConv2D(filters=32,kernel_size=(3,3),padding='same',activation='selu'))
model.add(tf.keras.layers.MaxPool2D())
model.add(tf.keras.layers.SeparableConv2D(filters=64,kernel_size=(3,3),padding='same',activation='selu'))
model.add(tf.keras.layers.GlobalAveragePooling2D())
model.add(tf.keras.layers.Dense(10,activation='softmax'))

保存模型

logdir = './cnn_selu_callbacks'
if not os.path.exists(logdir):
    os.mkdir(logdir)
output_model_file = os.path.join(logdir,"fashion_minist_model_1.h5")

设置自动下降学习率,早停

callbacks = [
    #tf.keras.callbacks.TensorBoard(log_dir=logdir), #log_dir将输出的日志保存在所要保存的路径中
    tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', 
                                                             patience=3, 
                                                             factor=0.5, 
                                                             min_lr=0.00001),
    #tf.keras.callbacks.ModelCheckpoint(output_model_file,save_best_only=True)
    tf.keras.callbacks.EarlyStopping(patience=6, min_delta=0.000001),
    tf.keras.callbacks.ModelCheckpoint(output_model_file,
                                       save_best_only=True,
                                       save_weights_only=True,monitor='val_loss')]

model.compile(optimizer='Adam',
              loss='sparse_categorical_crossentropy',
              metrics=['acc'])

history = model.fit(X_train, y_train,epochs=30,
                    shuffle=True,
                    batch_size=128,
                    validation_data=(X_test, y_test),
                    callbacks=callbacks)

Train on 60000 samples, validate on 10000 samples
Epoch 1/30
60000/60000 [==============================] - 11s 182us/sample - loss: 0.9925 - acc: 0.6357 - val_loss: 0.6348 - val_acc: 0.7541

Epoch 30/30
60000/60000 [==============================] - 5s 89us/sample - loss: 0.2477 - acc: 0.9084 - val_loss: 0.3021 - val_acc: 0.8947

XX = X_test[0].reshape(28,28)
XX.shape

(28, 28)

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 28, 28, 64)        640       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 14, 14, 64)        0         
_________________________________________________________________
separable_conv2d (SeparableC (None, 14, 14, 32)        2656      
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 7, 7, 32)          0         
_________________________________________________________________
separable_conv2d_1 (Separabl (None, 7, 7, 16)          816       
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 3, 3, 16)          0         
_________________________________________________________________
separable_conv2d_2 (Separabl (None, 3, 3, 32)          688       
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 1, 1, 32)          0         
_________________________________________________________________
separable_conv2d_3 (Separabl (None, 1, 1, 64)          2400      
_________________________________________________________________
global_average_pooling2d (Gl (None, 64)                0         
_________________________________________________________________
dense (Dense)                (None, 10)                650       
=================================================================
Total params: 7,850
Trainable params: 7,850
Non-trainable params: 0
_________________________________________________________________

可训练参数只有7850个,远远小于普通的卷积神经网络

plt.imshow(XX)

<matplotlib.image.AxesImage at 0x1b9f3c49400>

def plot_learning_curves(history):
    pd.DataFrame(history.history).plot(figsize = (8,5))
    plt.grid(True)
    plt.gca().set_ylim(0,1.2)
    plt.show()

plot_learning_curves(history)

分开画图:

def plot_learning_curves_1(history,label,ephochs,min_value,max_value,title):
    data = {}
    data[label] = history.history[label]
    data['val_'+label] = history.history['val_'+label]
    pd.DataFrame(data).plot(figsize = (8,5))
    plt.grid(True)
    plt.axis([0, ephochs, min_value, max_value])
    plt.xlabel('ephochs')
    plt.title(title,fontsize=20,fontweight='heavy')
    plt.show()

plot_learning_curves_1(history,'acc',30,0,1.2,title='acc')
plot_learning_curves_1(history,'loss',30,0,1,title='loss')

plt.plot(history.epoch,history.history.get(‘loss’),label=‘loss’,color=‘limegreen’,linewidth=‘2’)
plt.plot(history.epoch,history.history.get(‘val_loss’),label=‘val_loss’,color=‘tomato’,linewidth=‘2’)
plt.legend(loc=‘best’)
plt.title(“dasdasdas”)