kda lda matlab代码,机器学习初学代码（三） LDA（多分类）

# -*- coding: utf-8 -*-

# author: Xin Chen

import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

# 多分类通用:

# data 的格式应该是DataFrame的格式，其中前几列为特征，最后一列为标签

# label从0开始取得数字

def calWeights(data, K):

labels = np.unique(data['label'])

cNum = len(labels)

if K >= cNum:

print 'K is too large, please input again'

exit(0)

# Sw 类内散度矩阵

dim = np.shape(data)[1]-1 # 减1是因为有label

Sw = np.mat(np.zeros((dim, dim)))

for i in labels:

datai = data.ix[data['label'] == i]

Ci = np.mat(datai.drop('label', axis=1))

ui = np.mean(Ci, axis=0)

Si = (Ci-ui).T * (Ci-ui)

Sw += Si

# St 全局散度矩阵

C = np.mat(data.drop('label', axis=1))

u = np.mean(C, axis=0)

St = (C-u).T *(C-u)

# Sb 类间散度矩阵

Sb = St - Sw

S = Sw.I * Sb

eigValues, eigVectors = np.linalg.eig(S) # 用来求特征值和特征向量

order= eigValues.argsort()[::-1]

select = order[:K]

Weights = eigVectors[:, select]

return Weights

def LDAplot(data):

# 也要注意这个函数画的是投影点而不是原始的点

w = calWeights(data, K=2)

if np.shape(w)[1] != 2:

print '维度太大，无法画图！'

return

labels = np.unique(data['label'])

C = np.mat(data.drop('label', axis=1))

CC = C * w

corlist = ['r', 'y', 'g', 'b', 'w']

for i in labels:

CCi = CC[:, data['label'] == i]

plt.scatter(CCi[:, 0], CCi[:, 1], c=corlist[i], marker="o")

plt.show()

from sklearn import datasets

iris = datasets.load_iris()

X = iris.data

Y = iris.target

data = pd.DataFrame(X)

data['label'] = Y

Weights = calWeights(data, K=2)

print "opticalweights=", Weights

LDAplot(data)