logisticregression

from numpy import *
import random
import time
st = time.time()

def loaddata(filename):
    fr = open(''.join([filename, '.txt'])).readlines()
    trainx = [[1] + map(float, line.split()[:-1]) for line in fr] # trainx = [[1,12.2,22.4],[1,22.3,31.2],...]
    trainy = [[float(line.split()[-1])] for line in fr] # trainy = [0,1,1,0,...]
    return trainx, trainy

def sigmod(z):
    return 1.0 / (1 + exp(-z))

def optimizaion(trainx, trainy):
    trainxmat = mat(trainx)
    m = len(trainx)
    # beta = [0,0,0]
    beta = ones((len(trainx[0]),1)) # array
    # maxiter
    M = 500
    """
    # error permid
    e = 
    """
    """
    for i in xrange(M):
        #if error2sum > e:
        # z = betat.T * x = trainx (matricdoc)* beta = [beta.Tx1,beta.Tx2,...,beta.Txn]
        sigmodz = sigmod(trainxmat * beta)
        # [error_i = yi - sigmod(zi)]
        error = trainy - sigmodz
        # update beta
        beta += alpha * trainxmat.T * error
        print beta
        """
    # random gradascent
    for j in xrange(M):
        for i in xrange(m):
            # per span
            alpha = 0.01 + 4 / (1.0 + i +j)
            randid = random.randint(0, m - 1)
            sigmodz = sigmod(trainxmat[randid] * beta)
            error = trainy[randid] - sigmodz
            beta += alpha * trainxmat[randid].T * error
            #print beta

    return beta


def logregress(testx, beta):
    if mat(testx) * beta > 0: return [1.0]
    else: return [0.0]

def main():
    # step 1: loading data...
    print "step 1: loading data..."
    trainx, trainy = loaddata('horseColicTraining')
    testx, testy = loaddata('horseColicTest')
    """
    print 'trainx', trainx
    print 'trainy', trainy
    print 'testx', testx
    print 'testy', testy
    print 'testy[2]',testy[2]
    """

    # step 2: training...
    print "step 2: training..."
    beta = optimizaion(trainx, trainy)
    #print "beta = ",beta

    # step 3: testing...
    print "step 3: testing..."
    numTests = 10; errorSum = 0.0; l = len(testx)
    for j in xrange(numTests):
        errorcount = 0.0
        #print 'the total number is: ',l
        for i in xrange(l):
            if logregress(testx[i], beta) != testy[i]: 
                errorcount += 1
        #print "the number of error is: ", errorcount
        print "the error rate is: ", (errorcount / l)
        errorSum += (errorcount / l)
    print "after %d iterations the average error rate is: %f" %(numTests, errorSum/numTests)



"""
trainx, trainy = loaddata('testSet')
print trainy
optimizaion(trainx, trainy)
"""

main()

print "cost time: ", (time.time() - st)

""" lineregres
        # ssi = sigmod(zi) - sigmod(zi) ** 2
        ss = [sigmodzi - sigmodzi ** 2 for sigmodzi in sigmodz]
        # errssi = errori * ssi
        errss = map(lambda x, y: x * y, error, ss)
        # treri = errssi * trainxi(vector)
        trer = [errss[i] * array(trainx[i]) for i in xrange(m)]
        """