Perceptron emulation

perceptron.py

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from pandas import Series, DataFrame
import pandas.io.data as web

from numpy.random import randint, randn, rand, multivariate_normal

def run_perceptron(axes):
    n1 = randint(40,80)
    n2 = randint(40,80)
    mu1 = [rand()*10-5,rand()*10-5]
    mu2 = [rand()*10-5,rand()*10-5]

    cov1 = np.array([[rand()*3+1,0],[0,rand()*3+1]])
    theta = rand() * np.pi
    rot = np.array([[np.cos(theta),-np.sin(theta)],
                    [np.sin(theta), np.cos(theta)]])
    cov1 = np.dot(rot.T,np.dot(cov1,rot))

    cov2 = np.array([[rand()*3+1,0],[0,rand()*3+1]])
    theta = rand() * np.pi
    rot = np.array([[np.cos(theta),-np.sin(theta)],
                    [np.sin(theta), np.cos(theta)]])
    cov2 = np.dot(rot.T,np.dot(cov2,rot))

    df1 = DataFrame(multivariate_normal(mu1,cov1,n1),columns=['x','y'])
    df1['type']=1
    df2 = DataFrame(multivariate_normal(mu2,cov2,n2),columns=['x','y'])
    df2['type']=-1
    df = pd.concat([df1,df2],ignore_index=True)
    df = df.reindex(np.random.permutation(df.index))

    axes.clear()
    ymin, ymax = df.y.min(), df.y.max()
    xmin, xmax = df.x.min(), df.x.max()
    axes.set_ylim([ymin-1, ymax+1])
    axes.set_xlim([xmin-1, xmax+1])
    axes.scatter(df1.x, df1.y, marker='o')
    axes.scatter(df2.x, df2.y, marker='x')

    best = {'err_rate': 100}
    for c in range(5):
        dx, dy, d0 = rand()*2-1, rand()*2-1, rand()*2-1

        for i in range(40):
            for index, point in df.iterrows():
                x, y, type = point.x, point.y, point.type
                if type * (x * dx + y * dy + d0) >= 0:
                    dx -= type * x
                    dy -= type * y
                    d0 -= type * 1 
        err = 0
        for index, point in df.iterrows():
            x, y, type = point.x, point.y, point.type
            if type * (x * dx + y * dy + d0) >= 0:
                err += 1
        err_rate = err * 100 / len(df)
        if err_rate < best['err_rate']:
            best['err_rate'] = err_rate
            best['dx'] = dx
            best['dy'] = dy
            best['d0'] = d0
        if err_rate == 0:
            break

    dx, dy, d0 = best['dx'], best['dy'], best['d0']
    linex = np.arange(xmin-5, xmax+5)
    liney = - linex * dx / dy - d0 / dy
    label = "ERR: %.2f%%" % best['err_rate']
    axes.plot(linex,liney,label=label)
    axes.legend(loc='best')

if __name__ == '__main__':
    fig, axes = plt.subplots(2,2)
    positions = [[0,0],[0,1],[1,0],[1,1]]

    for pos in positions:
        run_perceptron(axes[pos[0],pos[1]])

    fig.show()