softmax(python实现)

最新推荐文章于 2024-07-22 21:49:01 发布

原创最新推荐文章于 2024-07-22 21:49:01 发布 · 881 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#python

machine learning 专栏收录该内容

9 篇文章

订阅专栏

本文介绍了一个多元逻辑回归算法的实现过程，包括数据加载、模型训练和预测等环节，并通过图表展示了损失函数的变化趋势，最终实现了0.96的高准确率。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

公式
这里写图片描述

实现

加载数据

def loadData():
    train_x = []
    train_y = []
    fileIn = open('testSet')
    for line in fileIn.readlines():
        lineArr = line.strip().split()
        train_x.append([1.0, float(lineArr[0]), float(lineArr[1])])
        train_y.append(float(lineArr[2]))
    return train_x, train_y

训练

def train(self, x, y):

        self.label_num = len(set(y))

        self.dataMat = np.mat(x)
        self.labelMat = np.mat(y).transpose()
        M, self.feature_num = np.shape(self.dataMat)
        self.theta = np.mat(np.ones((self.label_num, self.feature_num)))

        loss = []
        for kk in range(0, self.run_times):
            prob = self.condProb(self.dataMat)

            loss.append(self.cost(prob, y) / M)

            prob = -prob

            self.alpha = 0.001 + 0.1 * (self.run_times - kk) / self.run_times

            for m in range(M):
                prob[int(self.labelMat[m, 0]), m] += 1.0

            self.theta = self.theta + self.alpha * prob * self.dataMat / M

        import matplotlib.pyplot as plt

        plt.plot(xrange(self.run_times), loss)
        plt.show()


def condProb(self, x):
        prob = np.exp(self.theta * x.T)
        prob_sum = np.sum(prob, axis=0)
        prob_sum = prob_sum.repeat(self.label_num, axis=0)
        prob = prob / prob_sum
        return prob

def predict(self, x):
        probs = self.condProb(x)
        return probs


def cost(self, prob, y):
        costs = 0.0

        for i in xrange(len(y)):
            costs += np.log(prob[int(y[i]), i])
        return -costs