感知机对偶形式实现

import os
import numpy as np

# An example in that book, the training set and parameters' sizes are fixed
training_set = np.array([[[3, 3], 1], [[4, 3], 1], [[1, 1], -1]])

a = np.array([0, 0, 0])
b = 0
Gram = np.array([])
y = np.array(range(len(training_set))).reshape(1, 3)# used for labels
x = np.array(range(len(training_set)) * 2).reshape(3, 2)# used for features

# calculate the Gram matrix
def cal_gram():
    g = np.array([[0, 0, 0], [0, 0, 0], [0, 0, 0]])
    for i in range(len(training_set)):
        for j in range(len(training_set)):
            g[i][j] = np.dot(training_set[i][0], training_set[j][0])
    return g

# update parameters using stochastic gradient descent
def update(i):
    global a, b
    a[i] = a[i] + 1
    b = b + training_set[i][1]
    # print a, b # you can uncomment this line to check the process of stochastic gradient descent

# calculate the judge condition
def cal(key):
    global a, b, x, y
    i = 0
    for item in training_set:
        y[0][i] = item[1]
        i = i + 1
    temp = a * y
    res = np.dot(temp, Gram[key])
    res = (res + b) * training_set[key][1]
    return res[0]

# check if the hyperplane can classify the examples correctly
def check():
    global a, b, x, y
    flag = False
    for i in range(len(training_set)):
        if cal(i) <= 0:
            flag = True
            update(i)
    if not flag:
        i = 0
        for item in training_set:
            y[0][i] = item[1]
            x[i] = item[0]
            i = i + 1
        temp = a * y
        w = np.dot(temp, x)
        print "RESULT: w: " + str(w) + " b: "+ str(b)
        os._exit(0)
    flag = False
   

if __name__=="__main__":
    Gram = cal_gram()# initialize the Gram matrix
    for i in range(1000):
        check()
    print "The training_set is not linear separable. "

【メモ】svm実装

import numpy
from matplotlib import pyplot
import sys

def f(x, y):
    return x - y

if __name__ == '__main__':

    param = sys.argv

    numpy.random.seed()
    N = 30
    d = 2
    X = numpy.random.randn(N, d)
    T = numpy.array([1 if f(x, y) > 0 else - 1 for x, y in X])
    alpha = numpy.zeros(N)
    beta = 1.0
    eta_al = 0.0001 # update ratio of alpha
    eta_be = 0.1 # update ratio of beta
    itr = 1000

    for _itr in range(itr):
        for i in range(N):
            delta = 1 - (T[i] * X[i]).dot(alpha * T * X.T).sum() - beta * T[i] * alpha.dot(T)
            alpha[i] += eta_al * delta
        for i in range(N):
            beta += eta_be * alpha.dot(T) ** 2 / 2

    index = alpha > 0
    w = (alpha * T).T.dot(X)
    b = (T[index] - X[index].dot(w)).mean()

    if '-d' in param or '-s' in param:
        seq = numpy.arange(-3, 3, 0.02)
        pyplot.figure(figsize = (6, 6))
        pyplot.xlim(-3, 3)
        pyplot.ylim(-3, 3)
        pyplot.plot(seq, -(w[0] * seq + b) / w[1], 'k-')
        pyplot.plot(X[T ==  1,0], X[T ==  1,1], 'ro')
        pyplot.plot(X[T == -1,0], X[T == -1,1], 'bo')

        if '-s' in param:
            pyplot.savefig('graph.png')

        if '-d' in param:
            pyplot.show()

Postgresqlでto_chorを使用してintegerからString変換

■
select to_char(5, '09');

実行した結果は

_05

である。スペースが入れている。


■
select to_char(5, 'FM09');

実行した結果は

05

である。スペースが入れていない。

java日付比較時millisecond注意

Date dt3 = dateFormat("20161001", "yyyyMMdd");

Calendar cal = Calendar.getInstance();

cal.setTime(new Date(1475247600132L));
int year  = cal.get(Calendar.YEAR);
int month = cal.get(Calendar.MONTH);
int day   = 1;
cal.set(year, month, day, 0, 0, 0);
Date dt4 = cal.getTime();

System.out.println(dt3.getTime());    //出力：1475247600000

System.out.println(dt4.getTime());    //出力：1475247600132


Calendar calnoMillSecond = Calendar.getInstance();

calnoMillSecond.setTime(new Date(1475247600132L));
int year1  = cal.get(Calendar.YEAR);
int month1 = cal.get(Calendar.MONTH);
int day1   = 1;
calnoMillSecond.set(year1, month1, day1, 0, 0, 0);
calnoMillSecond.set(Calendar.MILLISECOND, 0);
Date dt5 = calnoMillSecond.getTime();

System.out.println(dt3.getTime());    //出力：1475247600000

System.out.println(dt5.getTime());    //出力：1475247600000

ミリ秒の0設定を注意。

讲座PPT

http://toritris.weebly.com/

http://www.massey.ac.nz/~mjjohnso/notes/59302/l10.html

https://cw.fel.cvut.cz/wiki/_media/courses/ae3m33ui/lectures/linclass-slides.pdf

http://wenku.baidu.com/link?url=V3cS4pGEbT0N1BUdgFtprvwVL7C5VpmLNbfk-TorMJAu_oXPi_jL7mNL23aYjEkGH4nFUDGvtTy062ROrOOVeiba0G8gjLNyPtEjIEtHyAK

java source資料
http://lcn.epfl.ch/tutorial/english/


https://datasciencelab.wordpress.com/2014/01/10/machine-learning-classics-the-perceptron/

https://appliedgo.net/perceptron/

https://www.reddit.com/r/MachineLearning/comments/15zrpp/please_explain_support_vector_machines_svm_like_i/

why sigmoid not step
http://fedc.wiwi.hu-berlin.de/fedc_homepage/xplore/tutorials/sfehtmlnode96.html

点到超平面计算
https://segmentfault.com/a/1190000005129669

http://blog.csdn.net/han_xiaoyang/article/details/50856583

Must Know Tips/Tricks in Deep Neural Networks

http://www.roguewave.com/getattachment/80bd5b86-ea29-457a-a0e1-db07e4cc5157/Neural-Networks-An-Introductiontarget=

https://mattmazur.com/2015/03/17/a-step-by-step-backpropagation-example/