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

"""

given a word and visualize near words

"""

import word2vec_boostpython as w2v

from sklearn.decomposition import PCA

import matplotlib.pyplot as plt

import matplotlib.font_manager

font = matplotlib.font_manager.FontProperties(fname='./ipag.ttc')

FONT_SIZE = 10

TEXT_KW = dict(fontsize=FONT_SIZE, fontweight='bold', fontproperties=font)

filename = 'word2vec/jawiki.bin'

#filename = 'word2vec/orj.bin'

print 'loading'

data = w2v.load(filename)

print 'loaded'

nbest = 15

while True:

query = raw_input('query: ')

if query.startswith('nbest='):

nbest = int(query[6:])

continue

if ', ' not in query:

words = [query] + w2v.search(data, query)[:nbest]

else:

words = query.split(', ')

print ', '.join(words)

mat = w2v.get_vectors(data)

word_indexes = [w2v.get_word_index(data, w) for w in words]

if word_indexes == [-1]:

print 'not in vocabulary'

continue

# do PCA

X = mat[word_indexes]

pca = PCA(n_components=2)

pca.fit(X)

print pca.explained_variance_ratio_

X = pca.transform(X)

xs = X[:, 0]

ys = X[:, 1]

# draw

plt.clf()

plt.scatter(xs, ys, marker = 'o')

for i, w in enumerate(words):

plt.annotate(

w.decode('utf-8', 'ignore'),

xy = (xs[i], ys[i]), xytext = (3, 3),

textcoords = 'offset points', ha = 'left', va = 'top',

**TEXT_KW)

plt.savefig('last.png')

print 'ok.'