tf.keras--DNN

深度神经网络

# DNN
import os
import sys
import time

import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import pandas as pd
import sklearn
import tensorflow as tf
from tensorflow import keras

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:]

Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz
32768/29515 [=================================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz
26427392/26421880 [==============================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz
8192/5148 [===============================================] - 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz
4423680/4422102 [==============================] - 0s 0us/step

# x = (x - u)/std  u:均值  std:标准差
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)

#构建分类训练模型  深度神经网络
model = keras.models.Sequential()

# 先将数据展开
model.add(keras.layers.Flatten(input_shape=[28, 28]))
# 循环定义深度网络层  20层
for _ in range(20):
    model.add(keras.layers.Dense(100, activation='relu'))
model.add(keras.layers.Dense(10, activation='softmax'))

model.compile(loss='sparse_categorical_crossentropy', 
             optimizer = 'sgd',
             metrics = ["accuracy"])
model.summary()

Model: "sequential_5"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
flatten_5 (Flatten)          (None, 784)               0         
_________________________________________________________________
dense_135 (Dense)            (None, 100)               78500     
_________________________________________________________________
dense_136 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_137 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_138 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_139 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_140 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_141 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_142 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_143 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_144 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_145 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_146 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_147 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_148 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_149 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_150 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_151 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_152 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_153 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_154 (Dense)            (None, 100)               10100     
_________________________________________________________________
dense_155 (Dense)            (None, 10)                1010      
=================================================================
Total params: 271,410
Trainable params: 271,410
Non-trainable params: 0
_________________________________________________________________

# 开始训练  返回值为训练过程中的历史数据, 可用于画图分析训练效果
history = model.fit(x_train_scaled, y_train, epochs=30, validation_data=(x_valid_scaled, y_valid))

Train on 55000 samples, validate on 5000 samples
Epoch 1/30
55000/55000 [==============================] - 6s 118us/sample - loss: 1.9670 - accuracy: 0.2617 - val_loss: 0.9742 - val_accuracy: 0.6188
Epoch 2/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.7580 - accuracy: 0.7104 - val_loss: 0.6848 - val_accuracy: 0.7546
Epoch 3/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.5655 - accuracy: 0.8069 - val_loss: 0.6270 - val_accuracy: 0.7654
Epoch 4/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.4782 - accuracy: 0.8374 - val_loss: 0.4817 - val_accuracy: 0.8352
Epoch 5/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.4337 - accuracy: 0.8534 - val_loss: 0.4426 - val_accuracy: 0.8494
Epoch 6/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.3987 - accuracy: 0.8640 - val_loss: 0.4086 - val_accuracy: 0.8660
Epoch 7/30
55000/55000 [==============================] - 6s 104us/sample - loss: 0.3730 - accuracy: 0.8731 - val_loss: 0.4299 - val_accuracy: 0.8546
Epoch 8/30
55000/55000 [==============================] - 6s 105us/sample - loss: 0.3541 - accuracy: 0.8791 - val_loss: 0.4203 - val_accuracy: 0.8578
Epoch 9/30
55000/55000 [==============================] - 6s 102us/sample - loss: 0.3367 - accuracy: 0.8845 - val_loss: 0.3803 - val_accuracy: 0.8724
Epoch 10/30
55000/55000 [==============================] - 6s 104us/sample - loss: 0.3268 - accuracy: 0.8870 - val_loss: 0.3742 - val_accuracy: 0.8736
Epoch 11/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.3048 - accuracy: 0.8939 - val_loss: 0.4097 - val_accuracy: 0.8588
Epoch 12/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.2938 - accuracy: 0.8968 - val_loss: 0.3568 - val_accuracy: 0.8788
Epoch 13/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.2828 - accuracy: 0.9005 - val_loss: 0.3840 - val_accuracy: 0.8720
Epoch 14/30
55000/55000 [==============================] - 6s 102us/sample - loss: 0.2718 - accuracy: 0.9039 - val_loss: 0.3409 - val_accuracy: 0.8834
Epoch 15/30
55000/55000 [==============================] - 6s 103us/sample - loss: 0.2648 - accuracy: 0.9059 - val_loss: 0.3693 - val_accuracy: 0.8722
Epoch 16/30
55000/55000 [==============================] - 6s 102us/sample - loss: 0.2552 - accuracy: 0.9105 - val_loss: 0.3517 - val_accuracy: 0.8804
Epoch 17/30
55000/55000 [==============================] - 6s 102us/sample - loss: 0.2494 - accuracy: 0.9127 - val_loss: 0.3709 - val_accuracy: 0.8780
Epoch 18/30
55000/55000 [==============================] - 6s 105us/sample - loss: 0.2425 - accuracy: 0.9147 - val_loss: 0.3405 - val_accuracy: 0.8854
Epoch 19/30
55000/55000 [==============================] - 6s 107us/sample - loss: 0.2354 - accuracy: 0.9164 - val_loss: 0.4019 - val_accuracy: 0.8762
Epoch 20/30
55000/55000 [==============================] - 6s 102us/sample - loss: 0.2294 - accuracy: 0.9197 - val_loss: 0.3553 - val_accuracy: 0.8804
Epoch 21/30
55000/55000 [==============================] - 6s 104us/sample - loss: 0.2221 - accuracy: 0.9215 - val_loss: 0.3425 - val_accuracy: 0.8790
Epoch 22/30
55000/55000 [==============================] - 6s 104us/sample - loss: 0.2183 - accuracy: 0.9220 - val_loss: 0.3460 - val_accuracy: 0.8892
Epoch 23/30
55000/55000 [==============================] - 6s 101us/sample - loss: 0.2147 - accuracy: 0.9237 - val_loss: 0.3653 - val_accuracy: 0.8820
Epoch 24/30
55000/55000 [==============================] - 6s 101us/sample - loss: 0.2084 - accuracy: 0.9270 - val_loss: 0.3359 - val_accuracy: 0.8916
Epoch 25/30
55000/55000 [==============================] - 6s 101us/sample - loss: 0.2026 - accuracy: 0.9272 - val_loss: 0.3620 - val_accuracy: 0.8868
Epoch 26/30
55000/55000 [==============================] - 6s 102us/sample - loss: 0.1946 - accuracy: 0.9299 - val_loss: 0.3868 - val_accuracy: 0.8848
Epoch 27/30
55000/55000 [==============================] - 6s 100us/sample - loss: 0.1932 - accuracy: 0.9303 - val_loss: 0.3561 - val_accuracy: 0.8942
Epoch 28/30
55000/55000 [==============================] - 6s 101us/sample - loss: 0.1906 - accuracy: 0.9319 - val_loss: 0.3564 - val_accuracy: 0.8902
Epoch 29/30
55000/55000 [==============================] - 6s 102us/sample - loss: 0.1827 - accuracy: 0.9343 - val_loss: 0.3716 - val_accuracy: 0.8818
Epoch 30/30
55000/55000 [==============================] - 6s 102us/sample - loss: 0.1794 - accuracy: 0.9364 - val_loss: 0.3587 - val_accuracy: 0.8868

def plot_learning_curves(history):
    pd.DataFrame(history.history).plot(figsize=(8, 5))
    plt.grid(True)
    plt.gca().set_ylim(0, 1)  
    plt.show()

plot_learning_curves(history)

model.evaluate(x_test_scaled, y_test)

10000/10000 [==============================] - 1s 61us/sample - loss: 0.4046 - accuracy: 0.8764





[0.40458174127936364, 0.8764]