Topic Modeling on NIPS Dataset Using Gaussian LDA w/ Word-Embeddings

In [86]:

    

import numpy as np
import os
from operator import itemgetter
from collections import Counter
import scipy.stats as stat
from gensim.models import Word2Vec
from nltk import corpus
import FastGaussianLDA2


    

In [97]:

    

wvmodel = Word2Vec.load_word2vec_format(
    "/Users/michael/Documents/Gaussian_LDA-master/data/glove.wiki/glove.6B.50d.txt", binary=False)
print "word-vector dimension: {}".format(wvmodel.vector_size())


    

    




WARNING:gensim.models.word2vec:consider setting layer size to a multiple of 4 for greater performance

In [108]:

    

wv_vocab = set(wvmodel.vocab.keys())
stops = set(corpus.stopwords.words(fileids="english"))


    

In [109]:

    

corpus = []
nips_path = "/Users/michael/Documents/GaussianLDA/data/"
for folder in os.listdir(nips_path)[1:]:
    for doc in os.listdir(nips_path + folder):
        with open(nips_path + folder + "/" + doc, 'r') as f:
            txt = f.read().split()
            txt = map(lambda x: x.lower(), txt)  # Lowercasing each word
            txt = filter(lambda word: [letter for letter in word if ord(letter) < 128 ], txt)  # Checking each word for ascci error
            txt = filter(lambda x: x not in stops, txt)  # Removing stop words
            txt = filter(lambda x: x.isalpha(), txt)  # Removing non-letter words (eg numbers and symbols)
            txt = filter(lambda x: len(x) > 2, txt)  # removing super short words and single letters
            txt = filter(lambda x: x in wv_vocab, txt)    
            txt = ' '.join(txt)
            corpus.append(txt)


    

In [ ]:

    

print "Number of documents in corpus: {}".format(len(corpus))


    

In [ ]:

    

reload(FastGaussianLDA2)
topics = 50
dim = 50
run_num = 1
outputfile = "/Users/michael/Documents/GaussianLDA/output/NIPS_{}_{}T_{}D_".format(str(run_num),
                                                                                  str(topics), 
                                                                                  str(dim))
lda = FastGaussianLDA2.Gauss_LDA(topics, corpus, word_vector_model=wvmodel, alpha=.5, outputfile=outputfile)
lda.fit(50)  # Number of samples to run


    

    




Done processing corpus with 111 documents
There are 9007 words that could be converted to word vectors in your corpus 
There are 0 words that could NOT be converted to word vectors
getting cluster centroids
[   425.   2899.    465.   1933.    560.   3618.   1019.    840.    752.
   4657.    413.   1823.    818.   2918.   2437.    626.   2464.   2074.
   5779.    579.   2841.    419.    451.    193.   4595.   2902.   2167.
   1466.    537.   6669.   1570.   2227.    720.   1901.    541.    934.
    494.   3562.    728.  10024.   9659.   3327.    446.   5769.   2181.
   3879.   3630.    732.    411.   4745.]
Initialization complete
Starting fit
print topic means
TOPIC 0: (u'lurching', u'wobbling', u'awkwardly', u'veering', u'momentarily', u'nimbly', u'gingerly', u'stealthily', u'yanking')
TOPIC 1: (u'consider', u'decision', u'considering', u'whether', u'accept', u'step', u'possibility', u'should', u'would')
TOPIC 2: (u'maximally', u'describable', u'axiomatically', u'biochemically', u'redundantly', u'advantageously', u'deterministically', u'contextualised', u'cognisant')
TOPIC 3: (u'text', u'reference', u'explaining', u'describing', u'read', u'articles', u'description', u'detailed', u'publish')
TOPIC 4: (u'trade-offs', u'sub-categories', u'inter-related', u'morphologies', u'derivations', u'subdomains', u'continuities', u'above-mentioned', u'polarities')
TOPIC 5: (u'the', u'part', u'also', u'as', u'country', u'both', u'member', u'in', u'united')
TOPIC 6: (u'moore', u'smith', u'collins', u'allen', u'anderson', u'freeman', u'thompson', u'walker', u'clark')
TOPIC 7: (u'infection', u'brain', u'inflammation', u'respiratory', u'disease', u'stomach', u'symptoms', u'ailments', u'treating')
TOPIC 8: (u'fine-tuning', u'easiness', u'pluralization', u'materialisation', u'spacial', u'signification', u'break-down', u'linearity', u'quantifying')
TOPIC 9: (u'many', u'other', u'are', u'mostly', u'number', u'some', u'most', u'mainly', u'several')
TOPIC 10: (u'souza', u'caballero', u'carvalho', u'osorio', u'almeida', u'barreto', u'coutinho', u'aranda', u'figueiredo')
TOPIC 11: (u'applies', u'permitting', u'requirement', u'apply', u'requiring', u'rules', u'requirements', u'provisions', u'applicable')
TOPIC 12: (u'cat', u'snake', u'yellow', u'dog', u'monkey', u'red', u'tree', u'pink', u'blue')
TOPIC 13: (u'evidence', u'investigators', u'investigation', u'examining', u'investigating', u'case', u'involved', u'revealed', u'investigated')
TOPIC 14: (u'studies', u'science', u'biology', u'psychology', u'studying', u'sciences', u'graduate', u'harvard', u'mathematics')
TOPIC 15: (u'ferina', u'jiwamol', u'kanoksilp', u'monechma', u'mangxamba', u'thongrung', u'el1l', u'pakad', u'rw96')
TOPIC 16: (u'taken', u'fire', u'carry', u'carried', u'ground', u'carrying', u'leaving', u'inside', u'immediately')
TOPIC 17: (u'sense', u'necessity', u'demonstrates', u'importance', u'determination', u'fundamental', u'genuine', u'respect', u'regard')
TOPIC 18: (u'finite', u'formula_1', u'discrete', u'linear', u'formula_3', u'inverse', u'formula_2', u'symmetric', u'formula_4')
TOPIC 19: (u'individual-level', u'affixation', u'caustics', u'foregrounded', u'technobabble', u'nonlinearities', u'm.sambucetti', u'sreerema', u'r.winters')
TOPIC 20: (u'shows', u'featured', u'feature', u'show', u'musical', u'music', u'features', u'featuring', u'original')
TOPIC 21: (u'subtle', u'expressive', u'idiosyncratic', u'playful', u'evocative', u'whimsical', u'startlingly', u'strikingly', u'wonderfully')
TOPIC 22: (u'honeywell', u'motorola', u'packard', u'xerox', u'lucent', u'3com', u'alliedsignal', u'emc', u'raytheon')
TOPIC 23: (u'chung', u'yu', u'ching', u'ying', u'yang', u'yong', u'lin', u'yi', u'wu')
TOPIC 24: (u'something', u'sure', u'how', u'so', u'what', u'always', u'really', u'anything', u'everyone')
TOPIC 25: (u'built-in', u'circuitry', u'utilizes', u'devices', u'interface', u'sensor', u'optimized', u'analog', u'utilizing')
TOPIC 26: (u'cost', u'payments', u'pay', u'cash', u'costs', u'funds', u'amount', u'money', u'benefits')
TOPIC 27: (u'horizontal', u'curved', u'rectangular', u'shaped', u'vertical', u'circular', u'cylindrical', u'surfaces', u'rows')
TOPIC 28: (u'hutagalung', u'bentivegna', u'chambless', u'quilici', u'vanags', u'pirkko', u'zety', u'frohlich', u'wilbourn')
TOPIC 29: (u'context', u'characteristics', u'concepts', u'aspects', u'examples', u'particular', u'forms', u'specific', u'distinct')
TOPIC 30: (u'excitation', u'amplitude', u'attenuation', u'oscillating', u'oscillations', u'electromagnetic', u'vibration', u'modulation', u'voltage')
TOPIC 31: (u'cause', u'persistent', u'impact', u'inevitably', u'caused', u'resulting', u'serious', u'avoid', u'consequence')
TOPIC 32: (u'sufficiently', u'inherently', u'problematic', u'incapable', u'overly', u'fairly', u'reasonably', u'socially', u'morally')
TOPIC 33: (u'proteins', u'molecules', u'inhibits', u'receptors', u'molecule', u'receptor', u'inhibition', u'protein', u'differentiation')
TOPIC 34: (u'richmond', u'carolina', u'michigan', u'maryland', u'ohio', u'virginia', u'connecticut', u'indiana', u'illinois')
TOPIC 35: (u'decoding', u'retrieval', u'templates', u'databases', u'synchronization', u'encrypting', u'template', u'query', u'typing')
TOPIC 36: (u'bsr', u'afic', u'qms', u'efm', u'dpg', u'cnu', u'oaa', u'gmi', u'lna')
TOPIC 37: (u'straight', u'third', u'second', u'break', u'behind', u'fourth', u'back', u'half', u'put')
TOPIC 38: (u'liquid', u'ingredients', u'mixture', u'milk', u'raw', u'butter', u'sugar', u'dried', u'fruit')
TOPIC 39: (u'systems', u'use', u'using', u'commercial', u'system', u'operating', u'technology', u'uses', u'equipment')
TOPIC 40: (u'fact', u'rather', u'though', u'yet', u'indeed', u'particular', u'very', u'this', u'although')
TOPIC 41: (u'rise', u'decline', u'higher', u'expectations', u'rate', u'growth', u'gains', u'drop', u'prices')
TOPIC 42: (u'lohm', u'iz', u'za', u'yekt', u'ber', u'yev', u'tz', u'buhs', u'noss')
TOPIC 43: (u'on', u'before', u'came', u'since', u'after', u'days', u'months', u'during', u'followed')
TOPIC 44: (u'stretches', u'area', u'along', u'adjacent', u'across', u'portion', u'crossing', u'northeast', u'east')
TOPIC 45: (u'improve', u'aim', u'focus', u'ensure', u'develop', u'enhance', u'necessary', u'creating', u'establish')
TOPIC 46: (u'cerebellum', u'brainstem', u'neurons', u'cortical', u'ganglia', u'cortex', u'thalamus', u'vestibular', u'axons')
TOPIC 47: (u'infer', u'imply', u'logically', u'implication', u'insofar', u'characterize', u'plausible', u'ascribe', u'discern')
TOPIC 48: (u'winkler', u'vogel', u'cleary', u'hauser', u'mendelsohn', u'pickard', u'weinberg', u'gottlieb', u'metzger')
TOPIC 49: (u'measured', u'depending', u'decreases', u'probability', u'ratio', u'varies', u'exceeds', u'decreasing', u'increases')


Document-Topic Counts:, [   425.   2899.    465.   1933.    560.   3618.   1019.    840.    752.
   4657.    413.   1823.    818.   2918.   2437.    626.   2464.   2074.
   5779.    579.   2841.    419.    451.    193.   4595.   2902.   2167.
   1466.    537.   6669.   1570.   2227.    720.   1901.    541.    934.
    494.   3562.    728.  10024.   9659.   3327.    446.   5769.   2181.
   3879.   3630.    732.    411.   4745.]
0 docs sampled
20 docs sampled
40 docs sampled

In [ ]:

    




    

In [ ]: