#encoding:latin-1
from py_nnma import *
import numpy as np
from pylab import *
import math, pprint, time


import cPickle

A = cPickle.load(file("documents.dat","rb"))

#
# A contains documents belonging to one of two topics.
# nnma of A is used to find membership of each document
#

m,n = A.shape

print "loaded document matrix with %d documents and %d terms" % (n, m)
print 


def NCW(X):
    """ nc weighted form of document matrix from: 
        
        "Document clustering based on non-negative matrix factorization"

        Wei Xu, Xin Liu, Yihong Gong 

        Proceedings of the 26th annual international ACM SIGIR conference on
        Research and development in informaion retrieval

        2003, pages 267 - 273

        http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.117.2293

    """
    m, n = X.shape

    XXTe = X.transpose()*X * np.ones(n,)
    XXTe[XXTe==0] = 1.0

    D = np.diag(XXTe**(-.5))
    D[np.isinf(D)] = 1.0
    D[np.isnan(D)] = 1.0

    return X * D



def scatt(c1, c2, t=""):

    """ scatter plot of vectors c1 and c2 
        
        colors are plottet corresponding to topic membership
        further a decision line is drawn
    """

    # first 114 docs are from "topic 1", the remaining are "topic 2"
    ndocs_topic_one = 114 

    offs = .0
    if sum(c1*c1):
        c1 /= np.linalg.norm(c1)

    if sum(c2*c2):
        c2 /= np.linalg.norm(c2)

    figure()
    title(t)
    for i, (v1,v2) in enumerate(zip(c1,c2)):
        if i < ndocs_topic_one:
            plot([v1+offs], [v2+offs],'r.')
        else:
            plot([v1+offs], [v2+offs],'g.')

    
    plot([0,max(c1)+offs],[0,max(c2)+offs], 'k:')
    axis([-.01, .2, -.01, .2])


print "bild nc weighted document matrix"

A=NCW(A)


print time.asctime()

B, C, obj,  _, _ = GDCLS_L1(A, 2, verbose=1)

print time.asctime()
print

scatt(C[0,:], C[1,:])

show()