TensorFlow2.0实现逻辑回归二分类，多分类

叁歲伎倆 2023-02-20 07:21 39阅读 0赞

代码路径：https://github.com/lilihongjava/leeblog\_python/tree/master/tensorflow\_logistic\_regression

数据集iris.csv，sklearn 鸢尾花数据集

## 二分类 ##

这里用二元的交叉熵作为二分类的损失函数，激活函数为sigmoid

layer0 = tf.keras.layers.Dense(1, input_shape=(x_data.shape[1],), activation='sigmoid')
     model = tf.keras.Sequential([layer0])
     model.compile(loss='binary_crossentropy', optimizer='adam')

## 多分类 ##

categorical\_crossentropy：多类的对数损失,它是一个多分类损失函数，可以配合着softmax一起使用。

layer0 = tf.keras.layers.Dense(class_num, input_shape=(x_data.shape[1],), activation='softmax')
     model = tf.keras.Sequential([layer0])
     model.compile(loss='categorical_crossentropy', optimizer='adam')

## 整体代码 ##

# encoding: utf-8
    """ @author: lee @time: 2020/6/24 16:35 @file: main.py @desc: """
    import pandas as pd
    import tensorflow as tf
    from tensorflow.python.keras.utils.np_utils import to_categorical
    
    from util.common_util import arg_check_transformation, multiple_gpu_strategy
    
    
    def model_builder(x_data, class_num):
        if class_num == 2:  # 逻辑回归二分类
            layer0 = tf.keras.layers.Dense(1, input_shape=(x_data.shape[1],), activation='sigmoid')
            model = tf.keras.Sequential([layer0])
            model.compile(loss='binary_crossentropy', optimizer='adam')  # 这里用二元的交叉熵作为二分类的损失函数
        else:  # 多分类
            layer0 = tf.keras.layers.Dense(class_num, input_shape=(x_data.shape[1],), activation='softmax')
            model = tf.keras.Sequential([layer0])
            model.compile(loss='categorical_crossentropy', optimizer='adam')
        return model
    
    
    def tf_logistic_regression(feature_column=None, label_column=None, class_num=None, gpu=None, input1=None,
                               output1=None):
        print("输入参数：", locals())
        feature_column = arg_check_transformation("list_name_str", "feature_column", feature_column)
        label_column = arg_check_transformation("list_name_str", "label_column", label_column)
        class_num = arg_check_transformation("int", "class_num", class_num)
        df = pd.read_csv(input1)
        try:
            x_data = df[feature_column]
            y_data = df[label_column]
            if class_num != 2 and y_data.shape[1] == 1:
                y_data = to_categorical(y_data)  # 一维的分类转成多列
                y_data = pd.DataFrame(y_data)
            if gpu:
                dataset, BATCH_SIZE, strategy = multiple_gpu_strategy(x_data, y_data)
                with strategy.scope():
                    model = model_builder(x_data, class_num)
                model.fit(dataset.batch(BATCH_SIZE), verbose=False)
            else:
                model = model_builder(x_data, class_num)
                model.fit(x_data, y_data, epochs=1000)
        except Exception:
            raise Exception("模型训练错误")
        print("模型训练完成")
        if output1:
            model.save(output1)
    
    
    if __name__ == '__main__':
        # 二分类
        tf_logistic_regression(feature_column="sepal length (cm),sepal width (cm),petal length (cm),petal width (cm)",
                               label_column="target", class_num=2,
                               gpu=False, input1="./data/iris_two.csv", output1="./data/output.h5")
        # 多分类
        tf_logistic_regression(feature_column="sepal length (cm),sepal width (cm),petal length (cm),petal width (cm)",
                               label_column="target", class_num=3,
                               gpu=False, input1="./data/iris.csv", output1="./data/output.h5")