import scipy,pylab,numpy
from numpy import *
from scipy import *
from pylab import *
from urllib import urlopen
from gzip import GzipFile
from scipy.spatial import distance
from scipy.spatial.distance import cdist

data = random_sample((100,2))
labels = (data[:,0]*0.7+data[:,1]*0.4>0.5)
sum(labels)

# plotting the data points
cla()
d0 = data[labels==False]
d1 = data[labels]
plot(d0[:,0],d0[:,1],"bo")
plot(d1[:,0],d1[:,1],"ro")
savefig("tmp.png")
print len(d0),len(d1)



cla()
d0 = data[labels==False]
d1 = data[labels]
xlim((0,1)); ylim((0,1))
plot(d0[:,0],d0[:,1],"bo")
plot(d1[:,0],d1[:,1],"ro")
plot([0.15,0.72],[1.0,0.0],"g") # guess
savefig("tmp.png")
print len(d0),len(d1)

# augmenting the original vectors
augmented = concatenate([ones((100,1)),data],axis=1)
flipped = augmented.copy()
flipped[labels==False] = -flipped[labels==False]

# random initialization of weight vector
a = random_sample((3))
a

# perceptron learning
for epoch in range(100):
    nchanged = 0
    for i in range(len(flipped)):
        if dot(a,flipped[i])>0: continue
        a += flipped[i]
        nchanged += 1
    print (epoch,nchanged),
    if nchanged==0: break
print

# result of learning
d,a0,a1 = a
print d,a0,a1

# plotting the learned decision boundary
cla()
d0 = data[labels==False]
d1 = data[labels]
xlim((0,1)); ylim((0,1))
plot(d0[:,0],d0[:,1],"bo")
plot(d1[:,0],d1[:,1],"ro")
plot([0,-d/a0],[-d/a1,0],"g")
savefig("tmp.png")
print len(d0),len(d1)