我尝试进行基础的机器学习。因此，这里是我用于二元分类的感知器类。

class perceptron():    def __init__(self, x, y, threshold=0.5, learning_rate=0.1, max_epochs=10):        self.threshold = threshold        self.learning_rate = learning_rate        self.x = x        self.y = y        self.max_epochs = max_epochs            def initialize(self):        self.weights = np.random.rand(len(self.x[0]))                    def train(self):        epoch = 0        while True:            error_count = 0            epoch += 1            for (x,y) in zip(self.x, self.y):                error_count += self.train_observation(x, y, error_count)            print('Epoch: {0} Error count: {1}'.format(epoch, error_count))            if error_count == 0:                print('Training successful')                break            if epoch >= self.max_epochs:                print('Reached max epochs')                break                    def train_observation(self, x, y, error_count):        result = np.dot(x, self.weights) > self.threshold        error = y - result        if error != 0:            error_count += 1            for index, value in enumerate(x):                self.weights[index] += self.learning_rate * error * value        return error_count            def predict(self, x):        return int(np.dot(x, self.weights) > self.threshold)

我想对列表值的总和进行分类，如果总和大于等于0（表示1），否则（表示0）。因此，我创建了50个长度为10的数组，每个数组包含随机整数值，范围为[-3, 3]：

def sum01(x):    if sum(x) >= 0:        return 1    else:        return 0x = np.random.randint(low=-3, high=3, size=(50,10))y = [sum01(z) for z in a]

然后我初始化并训练模型：

p = perceptron(x, y)p.initialize()p.train()

然后我进行检查，发现很多预测结果不正确，我做错了什么？

predics = [(p.predict(i), sumab(i)) for i in np.random.randint(low=-3, high=3, size=(10, 10))]print(predics)

回答：

在进行了小错误修复后重新运行你的代码，我看到损失减少到0，并且输出了正确的结果 –

p = perceptron(x, y)p.initialize()p.train()

Epoch: 1 Error count: 196608Epoch: 2 Error count: 38654836736Epoch: 3 Error count: 268437504Epoch: 4 Error count: 0Training successful

predics = [(p.predict(i), sum01(i)) for i in np.random.randint(low=-3, high=3, size=(10, 10))]print(predics)

[(1, 1), (0, 0), (0, 0), (0, 0), (1, 1), (0, 0), (0, 0), (0, 0), (0, 0), (0, 0)]

解决方案

你的代码需要进行几个快速修改 –

1. 定义x和y时：

x = np.random.randint(low=-3, high=3, size=(50,10))y = [sum01(z) for z in x] #将这里的a改为x

2. 获取预测结果时：

#将sumab改为sum01predics = [(p.predict(i), sum01(i)) for i in np.random.randint(low=-3, high=3, size=(10, 10))] 

这样应该就可以工作了。你的完整代码变为 –

class perceptron():    def __init__(self, x, y, threshold=0.5, learning_rate=0.1, max_epochs=10):        self.threshold = threshold        self.learning_rate = learning_rate        self.x = x        self.y = y        self.max_epochs = max_epochs            def initialize(self):        self.weights = np.random.rand(len(self.x[0]))                    def train(self):        epoch = 0        while True:            error_count = 0            epoch += 1            for (x,y) in zip(self.x, self.y):                error_count += self.train_observation(x, y, error_count)            print('Epoch: {0} Error count: {1}'.format(epoch, error_count))            if error_count == 0:                print('Training successful')                break            if epoch >= self.max_epochs:                print('Reached max epochs')                break                    def train_observation(self, x, y, error_count):        result = np.dot(x, self.weights) > self.threshold        error = y - result        if error != 0:            error_count += 1            for index, value in enumerate(x):                self.weights[index] += self.learning_rate * error * value        return error_count            def predict(self, x):        return int(np.dot(x, self.weights) > self.threshold)        def sum01(x):    if sum(x) >= 0:        return 1    else:        return 0    x = np.random.randint(low=-3, high=3, size=(50,10))y = [sum01(z) for z in x]p = perceptron(x, y)p.initialize()p.train()predics = [(p.predict(i), sum01(i)) for i in np.random.randint(low=-3, high=3, size=(10, 10))]print(predics)