Python Perceptron 不工作，似乎没有只使用 numpy 的库进行训练

问题描述

我不明白为什么我的感知器不起作用，它没有从训练数据中学习，我不确定为什么。每个时代都会出现错误并收敛，我在这里遗漏了什么吗？这是源代码。

import numpy as np
import matplotlib.pyplot as plt
 
trainData = np.genfromtxt('train.data',delimiter=",")
testData = np.genfromtxt('test.data',")
 
 
class Perceptron():
 
    def __init__(self,learnRate,maxEpoch):
        self.learnRate = learnRate
        self.maxEpoch = maxEpoch
 
    def train(self,data):
 
        np.random.seed(2)
        x0 = np.ones((len(data),1))
        w0 = np.zeros((1))
        weights = np.random.rand((len(data[0]))-1)
        weights = np.hstack([w0,weights])
        data = np.hstack([x0,data])
        np.random.shuffle(data)
        label = data[:,5:]
        data = data[:,:5]
        iter = 0
        for epoch in range(self.maxEpoch): #trainign
            errors = 0
            rowCount = 0
            for row in data:
                activation = np.inner(row,weights)
                y = label[rowCount]# Class label
                rowCount += 1
                if y * activation <= 0:
                    for wx in weights:
                        wx = wx + y * row[int(wx)]
                        weights[0] = weights[0] + y
                    errors += 1
            iter += 1
        print(f'Errors total: {errors}')
        return weights
 
 
def classPicker(classX,classY):
    set1 = trainData[:40,:4]
    set2 = trainData[40:80,:4]
    set3 = trainData[80:120,:4]
    class0 = np.full((40,1),-1)
    class1 = np.ones((40,1))
 
    if classX == 1 and classY == 2:
        return np.vstack([np.hstack([set1,class0]),np.hstack([set2,class1])])
    if classX == 2 and classY == 3:
        return np.vstack([np.hstack([set2,np.hstack([set3,class1])])
    if classX == 1 and classY == 3:
        return np.vstack([np.hstack([set1,class1])])
 
 
#Class 1 & 2
pececptron1 = Perceptron(1,20)
trainDataset = (classPicker(1,2))
pececptron1.train(trainDataset)
 
#Class 2 & 3
#pececptron2 = Perceptron(1,1)
#trainDataset = classPicker(2,3)
#pececptron2.train(trainDataset)
 
#Class 1 & 3
#pececptron3 = Perceptron(1,1)
#trainDataset = classPicker(1,3)
#pececptron3.train(trainDataset)

数据集是鸢尾花数据集的修改版本。数据数组在第 25 行看起来像这样，就在训练循环之前。

label = 40 x 1 array of 1 or -1 depending on class
weights = [0,0.4359949,0.02592623,0.54966248,0.43532239] first entry in all weights = w0 aka bias
data =  [1,5,3.4,1.5,0.2] first entry in all  = x0 (1) for the bias

解决方法

暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！

如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。

小编邮箱:dio#foxmail.com (将#修改为@）

numpy perceptron python