分类任务与激活函数

爱被打了一巴掌 2022-09-08 00:14 100阅读 0赞

> 分类任务可分为:二分类,多分类,多标签分类

#### 1.分类任务 ####

二分类  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70]  
多分类  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 1]  
多标签分类  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 2]

#### 2.建模上的区别 ####

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 3]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 4]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 5]

#### 3.代码实现 ####

# 激活函数练习
    
    # sigmoid
    import numpy as np
    
    
    # 输入的数字越小 越靠近0,数字越大越靠近1
    def sigmoid(x):
        return 1.0 / (1 + np.exp(-x))
    
    
    print(sigmoid(0.1))
    
    # 画图
    import matplotlib.pyplot as plt
    
    sigmoid_input = np.arange(-10, 10)
    sigmoid_output = sigmoid(sigmoid_input)
    print('sigmoid function input:{}'.format(sigmoid_input.shape))
    print('sigmoid function output:{}'.format(sigmoid_output.shape))
    plt.plot(sigmoid_input, sigmoid_output)
    plt.xlabel('sigmoid input')
    plt.xlabel('sigmoid output')
    plt.show()
    
    # softmax
    import math
    
    softmax_input = [1.0, 2.0, 3.0, 4.0, 5.0, 7.0, 8.0]
    z_exp = [math.exp(i) for i in softmax_input]
    print(z_exp)
    sum_z_exp = sum(z_exp)
    softmax_output = [round(ex/sum_z_exp,2) for ex in z_exp]
    print(softmax_input)
    print(softmax_output)
    
    plt.plot(softmax_input, softmax_output)
    plt.xlabel('softmax input')
    plt.xlabel('softmax output')
    plt.show()

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70]: /images/20220829/48a87e3e98a14e13b91cb53a6aae7347.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 1]: /images/20220829/acecca6552d34c2f928b3a50b4bbdc4c.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 2]: /images/20220829/0e094424aef14338bc09de9d18fe432a.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 3]: /images/20220829/0af4820336584899874acd39f16783c8.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 4]: /images/20220829/87d3091d25a4466396fcd990edbe5ae3.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0NvY2t0YWlsX3B5_size_16_color_FFFFFF_t_70 5]: /images/20220829/6ea8b33499624860b15ffbb50d14e654.png