【tensorflow】神经网络：自制数据集

在实际应用中，我们常常需要自制数据集，解决本领域应用，而数据通常是图片或文字，需要做格式转换，才能在训练时使用。

代码：

import tensorflow as tf
from PIL import Image
import numpy as np
import os

# 训练用的输入特征和标签
x_train_readpath = "class4/MNIST_FC/mnist_image_label/mnist_train_jpg_60000/"
y_train_readpath = "class4/MNIST_FC/mnist_image_label/mnist_train_jpg_60000.txt"
x_train_savapath = "class4/MNIST_FC/mnist_image_label/mnist_x_train.npy"
y_train_savapath = "class4/MNIST_FC/mnist_image_label/mnist_y_train.npy"

# 测试用的输入特征和标签
x_test_readpath = "class4/MNIST_FC/mnist_image_label/mnist_test_jpg_10000/"
y_test_readpath = "class4/MNIST_FC/mnist_image_label/mnist_test_jpg_10000.txt"
x_test_savapath = "class4/MNIST_FC/mnist_image_label/mnist_x_test.npy"
y_test_savapath = "class4/MNIST_FC/mnist_image_label/mnist_y_test.npy"

# 读取输入特征和标签
def generateData(x_path, y_path):
    f = open(y_path, "r")     # 以只读形式打开存放标签的文件
    contents = f.readlines()  # 按行读取文件中的所有数据
    f.close()                 # 关闭文件

    # 建立空列表，存放读出来的数据
    x, y = [], []
    for content in contents:
        # 数据存放形式为：文件名 标签
        # 以空格分开后，value[0]=文件名，value[1]=标签
        value = content.split()

        img_path = x_path + value[0]      # 拼接出训练图片完整路径
        img = Image.open(img_path)        # 读取图片
        img = np.array(img.convert("L"))  # 将图片变为 8位宽 灰度值的np.array格式
        img = img/255.0                   # 数据归一化

        x.append(img)                     # 保存读取出来的输入特征和标签
        y.append(value[1])

        print("loding:" + content)        # 打印状态提示

    x = np.array(x)         # [] -> np.array
    y = np.array(y)
    y = y.astype(np.int64)  # 将y中的数据统一设置为int64类型

    return x, y

if os.path.exists(x_train_savapath) and os.path.exists(y_train_savapath) and os.path.exists(x_test_savapath) and os.path.exists(y_test_savapath):
    # 数据文件已存在，直接读取
    x_train_save = np.load(x_train_savapath)
    x_train = np.reshape(x_train_save, (len(x_train_save), 28, 28))
    y_train = np.load(y_train_savapath)

    x_test_save = np.load(x_test_savapath)
    x_test = np.reshape(x_test_save, (len(x_test_save), 28, 28))
    y_test = np.load(y_test_savapath)
else:
    # 数据文件不存在，生成数据文件
    x_train, y_train = generateData(x_train_readpath, y_train_readpath)
    x_test, y_test = generateData(x_test_readpath, y_test_readpath)

    x_train_save = np.reshape(x_train, (len(x_train), -1))
    x_test_save = np.reshape(x_test, (len(x_test), -1))
    np.save(x_train_savapath, x_train_save)
    np.save(y_train_savapath, y_train)
    np.save(x_test_savapath, x_test_save)
    np.save(y_test_savapath, y_test)

# 声明神经网络结构
model = tf.keras.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation="relu"),
    tf.keras.layers.Dense(10, activation="softmax")
])

# 配置训练方法(优化器，损失函数，评测方法)
model.compile(optimizer=tf.keras.optimizers.Adam(),
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              metrics=[tf.keras.metrics.sparse_categorical_accuracy])

# 执行训练过程
model.fit(x_train, y_train,
          batch_size=32, epochs=5,
          validation_data=(x_test, y_test),
          validation_freq=1)

# 打印网络结构和参数
model.summary()