Tensorflow函数式API的使用

Posted 2020-12-13 Geek Song

在我们使用tensorflow时，如果不能使用函数式api进行编程，那么一些复杂的神经网络结构就不会实现出来，只能使用简单的单向模型进行一层一层地堆叠。如果稍微复杂一点，遇到了Resnet这种带有残差模块的神经网络，那么用简单的神经网络堆叠的方式则不可能把这种网络堆叠出来。下面我们来使用函数式API来编写一个简单的全连接神经网络：
首先导包：

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

导入图片数据集：mnist

(train_image,train_label),(test_image,test_label)=tf.keras.datasets.fashion_mnist.load_data()

归一化：

train_image=train_image/255
test_image=test_image/255#进行数据的归一化，加快计算的进程

搭建全连接神经网络：

input=keras.Input(shape=(28,28))
x=keras.layers.Flatten()(input)#调用input
x=keras.layers.Dense(32,activation="relu")(x)
x=keras.layers.Dropout(0.5)(x)#一层一层的进行调用上一层的结果
output=keras.layers.Dense(10,activation="softmax")(x)
model=keras.Model(inputs=input,outputs=output)
model.summary()

输出：

Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 28, 28)]          0         
_________________________________________________________________
flatten (Flatten)            (None, 784)               0         
_________________________________________________________________
dense (Dense)                (None, 32)                25120     
_________________________________________________________________
dropout (Dropout)            (None, 32)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 10)                330       
=================================================================
Total params: 25,450
Trainable params: 25,450
Non-trainable params: 0

拟合模型：

model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
              loss="sparse_categorical_crossentropy",
              metrics=[‘acc‘]
)
history=model.fit(train_image,
                  train_label,
                  epochs=15,
                  validation_data=(test_image,test_label))

输出：

Train on 60000 samples, validate on 10000 samples
Epoch 1/15
60000/60000 [==============================] - 4s 64us/sample - loss: 0.8931 - acc: 0.6737 - val_loss: 0.5185 - val_acc: 0.8160
Epoch 2/15
60000/60000 [==============================] - 3s 57us/sample - loss: 0.6757 - acc: 0.7508 - val_loss: 0.4805 - val_acc: 0.8230
Epoch 3/15
60000/60000 [==============================] - 3s 50us/sample - loss: 0.6336 - acc: 0.7647 - val_loss: 0.4587 - val_acc: 0.8369
Epoch 4/15
60000/60000 [==============================] - 3s 49us/sample - loss: 0.6174 - acc: 0.7689 - val_loss: 0.4712 - val_acc: 0.8294
Epoch 5/15
60000/60000 [==============================] - 3s 48us/sample - loss: 0.6080 - acc: 0.7732 - val_loss: 0.4511 - val_acc: 0.8404
Epoch 6/15
60000/60000 [==============================] - 3s 48us/sample - loss: 0.5932 - acc: 0.7773 - val_loss: 0.4545 - val_acc: 0.8407
Epoch 7/15
60000/60000 [==============================] - 3s 47us/sample - loss: 0.5886 - acc: 0.7772 - val_loss: 0.4394 - val_acc: 0.8428
Epoch 8/15
60000/60000 [==============================] - 3s 52us/sample - loss: 0.5820 - acc: 0.7788 - val_loss: 0.4338 - val_acc: 0.8506
Epoch 9/15
60000/60000 [==============================] - 3s 48us/sample - loss: 0.5742 - acc: 0.7839 - val_loss: 0.4393 - val_acc: 0.8454
Epoch 10/15
60000/60000 [==============================] - 3s 49us/sample - loss: 0.5713 - acc: 0.7847 - val_loss: 0.4422 - val_acc: 0.8477
Epoch 11/15
60000/60000 [==============================] - 3s 47us/sample - loss: 0.5642 - acc: 0.7858 - val_loss: 0.4325 - val_acc: 0.8488
Epoch 12/15
60000/60000 [==============================] - 3s 48us/sample - loss: 0.5582 - acc: 0.7873 - val_loss: 0.4294 - val_acc: 0.8492
Epoch 13/15
60000/60000 [==============================] - 3s 48us/sample - loss: 0.5574 - acc: 0.7882 - val_loss: 0.4263 - val_acc: 0.8523
Epoch 14/15
60000/60000 [==============================] - 3s 48us/sample - loss: 0.5524 - acc: 0.7888 - val_loss: 0.4350 - val_acc: 0.8448
Epoch 15/15
60000/60000 [==============================] - 3s 47us/sample - loss: 0.5486 - acc: 0.7901 - val_loss: 0.4297 - val_acc: 0.8493

最后验证集的精度达到了84%，这是一个仅仅使用全连接神经网络和softmax就能够得到的一个很不错的结果了！