逻辑回归-二分类

Posted 2020-10-16 McKay

class 二分类Demo
    {
        public static void Demo()
        {
            List<float[]> inputs_x = new List<float[]>();
            inputs_x.Add(new float[] { 0.9f, 0.6f });
            inputs_x.Add(new float[] { 2f, 2.5f });
            inputs_x.Add(new float[] { 2.6f, 2.3f });
            inputs_x.Add(new float[] { 2.7f, 1.9f });

            List<float> inputs_y = new List<float>();
            inputs_y.Add(0);
            inputs_y.Add(1);
            inputs_y.Add(1);
            inputs_y.Add(1);

            float[] weights = new float[inputs_x.First().Length+1];//加上b
            for (var i = 0; i < weights.Length; i++)
                weights[i] = (float)new Random().NextDouble();

            int epoch = 30000;
            float epsilon = 0.00000000000000001f;
            float lr = 0.01f;

            float lastCost = 0;

            for (var epoch_i = 0; epoch_i <= epoch; epoch_i++)
            {
                //随机获取input
                var batch = GetRandomBatch(inputs_x, inputs_y, 2);

                float[] weights_in_poch = new float[weights.Length];

                foreach (var x_y in batch)
                {
                    var x1 = x_y.Item1.First();
                    var x2 = x_y.Item1.Skip(1).Take(1).First();
                    var target_y = x_y.Item2;

                    float diffWithTargetY = target_y - Sigmoid(fun(x1, x2, weights[1], weights[2], weights[0]));

                    weights_in_poch[0] += diffWithTargetY * dy_b(x1, x2);
                    weights_in_poch[1] += diffWithTargetY * dy_theta1(x1, x2);
                    weights_in_poch[2] += diffWithTargetY * dy_theta2(x1, x2);
                }

                for (var i = 0; i < weights.Length; i++)
                    weights[i] += lr * weights_in_poch[i];

                float totalErrorCost = 0f;
                foreach (var x_y in batch)
                {
                    var x1 = x_y.Item1.First();
                    var x2 = x_y.Item1.Skip(1).Take(1).First();
                    var target_y = x_y.Item2;

                    float diffWithTargetY = target_y - Sigmoid(fun(x1, x2, weights[1], weights[2], weights[0]));
                    totalErrorCost += (float)System.Math.Pow(diffWithTargetY, 2) / 2;
                }

                float cost = totalErrorCost / batch.Count;

                if (System.Math.Abs(cost - lastCost) <= epsilon)
                {
                    Console.WriteLine(string.Format("EPOCH {0}", epoch_i));
                    Console.WriteLine(string.Format("LAST MSE {0}", lastCost));
                    Console.WriteLine(string.Format("MSE {0}", cost));
                    break;
                }

                lastCost = cost;

                if (epoch_i % 100 == 0 || epoch_i == epoch)
                {
                    Console.WriteLine(string.Format("MSE {0}", cost));
                }
            }

            print(weights[1], weights[2], weights[0]);

            Console.ReadLine();
        }

        private static List<Tuple<float[], float>> GetRandomBatch(List<float[]> inputs_x, List<float> inputs_y, int maxCount)
        {
            List<Tuple<float[], float>> lst = new List<Tuple<float[], float>>();

            System.Random rnd = new Random((int)DateTime.Now.Ticks);

            int count = 0;
            while (count < maxCount)
            {
                int rndIndex = rnd.Next(inputs_x.Count);
                var item = Tuple.Create<float[], float>(inputs_x[rndIndex], inputs_y[rndIndex]);
                lst.Add(item);
                count++;
            }

            return lst;
        }

        private static void print(float theta1, float theta2, float b)
        {
            Console.WriteLine(string.Format("y=sigmoid({0}*x1+{1}*x2+{2})", theta1, theta2, b));
        }

        private static float Sigmoid(double x)
        {
            double one = 1;
            var result= one / (one + System.Math.Exp(-x));
            return (float)result;
        }

        private static float fun(float x1, float x2, float theta1, float theta2, float b)
        {
            return theta1 * x1 + theta2 * x2 + b;
        }
        private static float dy_theta1(float x1, float x2)
        {
            return x1;
        }

        private static float dy_theta2(float x1, float x2)
        {
            return x2;
        }

        private static float dy_b(float x1, float x2)
        {
            return 1;
        }
    }

 

 

import matplotlib.pyplot as plt
import numpy as np

x1=np.array([0.9,2,2.6,2.7], dtype=np.float)
x2=np.array([0.6,2.5,2.3,1.9], dtype=np.float)


def sigmoid(v):
    return 1/ (1 + np.exp(-v))

#y=sigmoid(0.8277054*x1+0.893053*x2+-0.7201675)
#y=sigmoid(1.968242*x1+4.206787*x2+-7.930489)
def y_function(x1, x2):
    #return sigmoid(0.8277054*x1+0.893053*x2-0.7201675)
    return sigmoid(1.968242*x1+4.206787*x2-7.930489)

y=y_function(x1, x2)

for index, y_value in enumerate(y):
    if(y_value>0.5):
        plt.scatter([x1[index]], [x2[index]], c = \'red\',marker = \'o\')
    else:
        plt.scatter([x1[index]], [x2[index]], c = \'blue\',marker = \'o\')

plt.show()

 

3D图形：

import matplotlib.pyplot as plt
import numpy as np
from mpl_toolkits.mplot3d import Axes3D

def sigmoid(v):
    return 1/ (1 + np.exp(-v))

x1=np.linspace(1, 5, 100)
x2=np.linspace(1, 5, 100)
y=1.968242*x1+4.206787*x2-7.930489
y_s=sigmoid(y)


fig = plt.figure()
ax = Axes3D(fig)
ax.scatter(x1, x2, y_s, c=\'r\', marker=\'.\', s=50, label=\'\')
plt.show()