Doc2Vec

In [1]:

    

import re
import random
import nltk.data
import numpy as np
import pandas as pd
from bs4 import BeautifulSoup
from nltk.corpus import stopwords
from gensim.models import Doc2Vec
from gensim.models.doc2vec import LabeledSentence
from sklearn.ensemble import RandomForestClassifier


    

    




C:\Users\anton\Anaconda2\lib\site-packages\gensim\utils.py:840: UserWarning: detected Windows; aliasing chunkize to chunkize_serial
  warnings.warn("detected Windows; aliasing chunkize to chunkize_serial")
C:\Users\anton\Anaconda2\lib\site-packages\gensim\utils.py:1015: UserWarning: Pattern library is not installed, lemmatization won't be available.
  warnings.warn("Pattern library is not installed, lemmatization won't be available.")

In [2]:

    

#Loading the differents sets of data.
train = pd.read_csv( "labeledTrainData.tsv", header=0, 
 delimiter="\t", quoting=3, encoding="utf-8" )
test = pd.read_csv( "testData.tsv", header=0, delimiter="\t", quoting=3, encoding="utf-8" )
unlabeled_train = pd.read_csv( "unlabeledTrainData.tsv", header=0, 
 delimiter="\t", quoting=3, encoding="utf-8" )


    

In [3]:

    

#Getting the reviews for each set of data.
x_train = train['review']
x_test = test['review']
x_unlabeled = unlabeled_train['review']

#Getting the sentiment just for train. In unlabeled_train we don't
#have sentiment column as same in test.
y_train = train['sentiment']


    

In [4]:

    

def review_to_wordlist( review, remove_stopwords=False ):
    # Function to convert a document to a sequence of words,
    # optionally removing stop words.  Returns a list of words.
    #
    clean_reviews = []
    for i in range(len(review)):
    # 1. Remove HTML
        review_text = BeautifulSoup(review[i]).get_text()
    #  
    # 2. Remove non-letters
        review_text = re.sub("[^a-zA-Z]"," ", review_text)
    #
    # 3. Convert words to lower case and split them
        words = review_text.lower().split()
    #
    # 4. Optionally remove stop words (false by default)
        if remove_stopwords:
            stops = set(stopwords.words("english"))
            words = [w for w in words if not w in stops]
    #
        clean_reviews.append(words)
    # 5. Return a list of reviews
    return(clean_reviews)


    

In [5]:

    

#Cleaning the data with the reviews in each set of data
x_train = review_to_wordlist(x_train)
x_test = review_to_wordlist(x_test)
x_unlabeled = review_to_wordlist(x_unlabeled)


    

    




C:\Users\anton\Anaconda2\lib\site-packages\bs4\__init__.py:181: UserWarning: No parser was explicitly specified, so I'm using the best available HTML parser for this system ("lxml"). This usually isn't a problem, but if you run this code on another system, or in a different virtual environment, it may use a different parser and behave differently.

The code that caused this warning is on line 174 of the file C:\Users\anton\Anaconda2\lib\runpy.py. To get rid of this warning, change code that looks like this:

 BeautifulSoup([your markup])

to this:

 BeautifulSoup([your markup], "lxml")

  markup_type=markup_type))

In [6]:

    

def labelizeReviews(reviews, label_type):
    labelized = []
    for i,v in enumerate(reviews):
        label = '%s_%s'%(label_type,i)
        labelized.append(LabeledSentence(v, [label]))
    return labelized


    

In [7]:

    

x_concatenate = np.concatenate((x_train, x_unlabeled))


    

In [8]:

    

#Doing labilize procces to each data set, specifying in the parameter "label_type"
#what kind of data is the given data. 
x_train2 = labelizeReviews(x_concatenate, 'TRAIN')
x_test2 = labelizeReviews(x_test, 'TEST')


    

In [9]:

    

size = 50
model= Doc2Vec(min_count=1, window=5, size=size, sample=1e-4, negative=5, workers=8)


    

In [10]:

    

#build vocab over all reviews
model.build_vocab(x_train2)


    

In [11]:

    

import random as ra


    

In [12]:

    

ra.seed(12345)
alpha = 0.025
min_alpha = 0.001
num_epochs = 5
alpha_delta = (alpha - min_alpha) / num_epochs

for epoch in range(num_epochs):
    print(epoch)
    ra.shuffle(x_train2)
    model.alpha = alpha
    model.min_alpha = alpha
    model.train(x_train2)
    alpha -= alpha_delta


    

    




0
1
2
3
4

In [13]:

    

#Infering over the test reviews in base the previous model built
test_vectors = np.zeros((len(x_test),size))
for i in range(len(x_test)):
    test_vectors[i] = model.infer_vector(x_test[i])


    

In [ ]:

    

#Getting the train vectors from the model with the function docvecs. 
#Remember that when we concatenate train and unlabeled sets, the first 
#25,000 rows are from train set. The rest 50,000 rows are from the unlabeled set.
train_vectors = np.zeros((len(x_train), size))
for i in range(len(x_train)):
    train_vectors[i] = model.docvecs[i]


    

In [15]:

    

#model.save('my_model.doc2vec')


    

In [16]:

    

# Fit a random forest and extract predictions 
forest = RandomForestClassifier(n_estimators = 100)
# Fitting the forest may take a few minutes
print "Fitting a random forest to labeled training data..."
forest = forest.fit(train_vectors,train["sentiment"])


    

    




Fitting a random forest to labeled training data...

In [17]:

    

results = forest.predict(test_vectors)


    

In [18]:

    

output = pd.DataFrame(data={"id":test["id"], "sentiment":results})
output.to_csv( "Doc2Vec.csv", index=False, quoting=3 )


    

In [20]:

    

from sklearn.linear_model import LogisticRegression


    

In [21]:

    

lgr = LogisticRegression()
lgr = lgr.fit(train_vectors, train["sentiment"])


    

In [22]:

    

results = lgr.predict(test_vectors)


    

In [23]:

    

output = pd.DataFrame(data={"id":test["id"], "sentiment":results})
output.to_csv( "Doc2Vec_lgr.csv", index=False, quoting=3 )


    

In [ ]: