TensorFlow CNN mnist

手写字体识别

tensorflow2.6.0 

import tensorflow as tf
from tensorflow.keras import datasets
from tensorflow.keras import Model
from tensorflow import keras
import numpy as np
mnist=tf.keras.datasets.mnist
(x_train,y_train),(x_test,y_test)=mnist.load_data()
# 图片数据除以255，将值规整在（0-1），数据标准化
x_train,x_test=x_train/255.,x_test/255.
# 增加一个维度（28,28,1）
x_train=tf.expand_dims(x_train,-1) 
x_test=tf.expand_dims(x_test,-1)
# onehot
y_train=np.float32(tf.keras.utils.to_categorical(y_train,num_classes=10))
y_test=np.float32(tf.keras.utils.to_categorical(y_test,num_classes=10))
batch_size=512
train_dataset=tf.data.Dataset.from_tensor_slices((x_train,y_train)).batch(batch_size).shuffle(batch_size*10)
test_dataset=tf.data.Dataset.from_tensor_slices((x_test,y_test)).batch(batch_size)
input_xs=tf.keras.Input([28,28,1])
conv=tf.keras.layers.BatchNormalization()(input_xs)
conv=tf.keras.layers.Conv2D(32,3,padding='SAME',activation=tf.nn.relu)(conv)
conv=tf.keras.layers.Conv2D(64,3,padding='SAME',activation=tf.nn.relu)(conv)
conv=tf.keras.layers.MaxPool2D(strides=[1,1])(conv)
conv=tf.keras.layers.Conv2D(128,3,padding='SAME',activation=tf.nn.relu)(conv)
flat=tf.keras.layers.Flatten()(conv)
dense=tf.keras.layers.Dense(512,activation=tf.nn.relu)(flat)
logits=tf.keras.layers.Dense(10,activation=tf.nn.softmax)(dense)
model=tf.keras.Model(inputs=input_xs,outputs=logits)
print(model.summary())
model.compile(optimizer=tf.keras.optimizers.Adam(1e-3),
loss=tf.keras.losses.categorical_crossentropy,
metrics=['accuracy'])
model.fit(train_dataset,epochs=10)
score=model.evaluate(test_dataset)
print('last score:',score)

Model: "model_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_4 (InputLayer)         [(None, 28, 28, 1)]       0         
_________________________________________________________________
batch_normalization_3 (Batch (None, 28, 28, 1)         4         
_________________________________________________________________
conv2d_9 (Conv2D)            (None, 28, 28, 32)        320       
_________________________________________________________________
conv2d_10 (Conv2D)           (None, 28, 28, 64)        18496     
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 27, 27, 64)        0         
_________________________________________________________________
conv2d_11 (Conv2D)           (None, 27, 27, 128)       73856     
_________________________________________________________________
flatten_3 (Flatten)          (None, 93312)             0         
_________________________________________________________________
dense_6 (Dense)              (None, 512)               47776256  
_________________________________________________________________
dense_7 (Dense)              (None, 10)                5130      
=================================================================
Total params: 47,874,062
Trainable params: 47,874,060
Non-trainable params: 2
_________________________________________________________________
None
Train on 118 steps
Epoch 1/10
118/118 [==============================] - 337s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.3205 - acc: 0.9263
Epoch 2/10
118/118 [==============================] - 344s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0432 - acc: 0.9866
Epoch 3/10
118/118 [==============================] - 339s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0248 - acc: 0.9923
Epoch 4/10
118/118 [==============================] - 332s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0181 - acc: 0.9943
Epoch 5/10
118/118 [==============================] - 325s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0123 - acc: 0.9964
Epoch 6/10
118/118 [==============================] - 329s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0091 - acc: 0.9974
Epoch 7/10
118/118 [==============================] - 323s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0065 - acc: 0.9982
Epoch 8/10
118/118 [==============================] - 322s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0057 - acc: 0.9986
Epoch 9/10
118/118 [==============================] - 329s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0054 - acc: 0.9987
Epoch 10/10
118/118 [==============================] - 325s 3s/step - batch: 58.5000 - size: 1.0000 - loss: 0.0041 - acc: 0.9992

last score: [0.03538419213145971, 0.9914]