%pylab inline
%load_ext autoreload
%autoreload 2

import pandas
import sys

# specific imports
sys.path += ['../scripts/']
from utils import *
from pyx import *
from wand.image import Image as WImage
from def_visualization import *

word_list = read_words('./vocab.dat')
experiment_dir = '../experiments/def_wikipedia_1434725288667'

t = map(softrect, load_bin_model(experiment_dir + '/train_iter01000.model.bin'))

W0_shape, W0_scale, z0_shape, z0_scale = t[:4]    
z1_shape, z1_scale, z2_shape, z2_scale = t[4:8]        
W1_shape, W1_scale, W2_shape, W2_scale = t[8:]
W0_mean = W0_shape * W0_scale
W1_mean = W1_shape * W1_scale
W2_mean = W2_shape * W2_scale

g = top_words(W0_mean, word_list, k=10, W_shape=W0_shape, show_weight=True)

#For each group we show: 
# 1) Most probable words
# 2) Top topics per group including their probability
# 3) The top word per topic
g = top_groups(W1_mean, W0_mean, word_list, k1=3, k=10, show_weight=True)

g = top_supers(W2_mean, W1_mean, W0_mean, word_list, k2=5, k1=5, k=6, show_weight=True)