人工神经网络

二、人工神经网络

监督学习问题分为两类：分类(输出分类标签)和回归(输出连续数字)，这些都能通过线性或者非线性模型来解决，ANN的基本构建块就是线性模型。

代码如下（示例）：

import numpy as np

def activation_function(inpt):
    if(inpt>250):
        return 250
    else:
        return inpt

def prediction_error(desired,expected):
    return np.abs(np.mean(desired-expected))

def update_weight(weights,predicted,idx):
    weights=weights+0.00001*(desired_output[idx]-predicted)*inputs[idx]
    return weights

weights=np.array([0.05,.1])
inputs=np.array([60,40,100,300,-50,310])
desired_output=np.array([60,40,150,250,-50,250])

def training_loop(inpt,weights):
    error=1
    idx=0
    iteration=0
    while(iteration<2000 or error>=0.01):
        predicted=activation_function(weights[0]+weights[1]*inputs[idx])
        error=prediction_error(desired_output[idx],predicted)
        weights=update_weight(weights,predicted,idx)
        idx=idx+1
        idx=idx%inputs.shape[0]
        iteration=iteration+1
    return error,weights

error,new_weights=training_loop(inputs,weights)
print('----------Final Results----------')
print('Learned Weights:',new_weights)
new_inputs=np.array([10,240,550,-160])
new_outputs=np.array([10,240,250,-160])
for i in range(new_inputs.shape[0]):
    print('Sample ',i+1,', Expected=',new_outputs[i],', Predicted=',activation_function(new_weights[0]+new_weights[1]*new_inputs[i]))