TensorFlow实战——图像分类

import numpy as np
import tensorflow as tf
from tensorflow import keras
import pandas as pd
import matplotlib.pyplot as plt
print(tf.__version__)
print(keras.__version__)

# 加载数据集
fashion_mnist = keras.datasets.fashion_mnist
(X_train_full, y_train_full), (X_test, Y_test)=fashion_mnist.load_data()
# 查看训练集的形状和数据类型
print(X_train_full.shape, X_train_full.dtype)
# 比例缩放和像素强度降低到0-1,创建一个验证集
X_valid, X_train = X_train_full[:5000]/255.0, X_train_full[5000:]/255.0
Y_valid, Y_train = y_train_full[:5000], y_train_full[5000:]
# 设置类名列表
class_names=["T_shirt/top", "Trouser", "Pullover", "Dress", "Coat", "Sandal", "shirt","Sneaker"
             , "Bag", "Anlle boot"]
# 查看第一幅图
print(class_names[Y_train[0]])

# 搭建网络模型
model = tf.keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28, 28]))
model.add(keras.layers.Dense(300, activation="relu"))
model.add(keras.layers.Dense(100, activation="relu"))
model.add(keras.layers.Dense(10, activation="softmax"))# 输出十个概率分布，看属于哪一个
model.compile(loss="sparse_categorical_crossentropy", optimizer="sgd", metrics=["accuracy"])
history=model.fit(X_train, Y_train, epochs=5, validation_data=(X_valid,Y_valid))
#print(history)

# 绘制history曲线
pd.DataFrame(history.history).plot(figsize=(8,5))
plt.grid(True)
plt.gca().set_ylim(0,1)
plt.show()


# 在测试集上测试
print(model.evaluate(X_test, Y_test))
#仅使用测试集的前三个例子
X_new=X_test[:3]
Y_proba= model.predict(X_new)
print(Y_proba.round(2))
y_pre=model.predict(X_new)
print(y_pre)

自动下载图像数据集；

一、Keras的mnist数据集
mnist数据集是KerasAPI公开的数据集，是（28，28）的图像数据集

二、建立sequential顺序model
对于分类问题必须使用softmax作为输出层的激活函数；如果每一个实例的每个类都有一个目标概率，即使用one—hot编码，在loss的设置时使用categorical_crossentropy；而下面的代码是基于稀疏标签（每个类只有一个目标类索引），并且类间互斥，则设置loss="sparse_categorical_crossentropy"
 

运行结果：

Epoch 1/5
1719/1719 [==============================] - 2s 1000us/step - loss: 0.7156 - accuracy: 0.7644 - val_loss: 0.5408 - val_accuracy: 0.8104
Epoch 2/5
1719/1719 [==============================] - 2s 968us/step - loss: 0.4902 - accuracy: 0.8283 - val_loss: 0.4420 - val_accuracy: 0.8544
Epoch 3/5
1719/1719 [==============================] - 2s 944us/step - loss: 0.4447 - accuracy: 0.8434 - val_loss: 0.4368 - val_accuracy: 0.8472
Epoch 4/5
1719/1719 [==============================] - 2s 941us/step - loss: 0.4176 - accuracy: 0.8533 - val_loss: 0.4170 - val_accuracy: 0.8572
Epoch 5/5
1719/1719 [==============================] - 2s 955us/step - loss: 0.3969 - accuracy: 0.8604 - val_loss: 0.3848 - val_accuracy: 0.8712

绘制准确率和验证集上的准确率