ravishchawla

# Obtain predictions
predictions = model.predict_classes(X_test)

# Convert Y_test to the same format as predictions
actuals = [0 if y[0] == 1 else 1 for y in Y_test];

# Use SkLearn's Metrics module
accuracy_score(predictions, actuals)

ravishchawla

embeddings_index = dict();
with open('data/glove.6B.100d.txt') as f:
    for line in f:
        values = line.split();
        word = values[0];
        coefs = np.asarray(values[1:], dtype='float32');
        embeddings_index[word] = coefs;
        
vocab_size = len(sequence_dict);
embeddings_matrix = np.zeros((vocab_size, 100));

ravishchawla

model = Sequential();
model.add(Embedding(len(word_dict), max_cap, input_length=max_cap));
model.add(LSTM(60, return_sequences=True, recurrent_dropout=0.5));
model.add(Dropout(0.5))
model.add(LSTM(60, recurrent_dropout=0.5));
model.add(Dense(60, activation='relu'));
model.add(Dense(2, activation='softmax'));
print(model.summary());

optimizer = Adam(lr=0.01, decay=0.001);

ravishchawla

model = Sequential();
model.add(Embedding(len(word_dict), max_cap, input_length=max_cap));
model.add(LSTM(100, return_sequences=True));
model.add(LSTM(100));
model.add(Dense(100, activation='relu'));
model.add(Dense(2, activation='softmax'));
print(model.summary());

optimizer = Adam(lr=0.001, decay=0.0001);
model.compile(loss='binary_crossentropy', optimizer=optimizer, metrics=['accuracy'])

ravishchawla

# Truncate and Pad reviews at a Maximum cap of 60 words.
max_cap = 60;
X = pad_sequences(reviews_encoded, maxlen=max_cap, truncating='post')

# Obtain a One-hot Y array for each review label.
Y = np.array([[0,1] if '0' in label else [1,0] for label in labels])

# Get a randomized sequence of positions to shuffle reviews
np.random.seed(1024);
random_posits = np.arange(len(X))

ravishchawla

# Use a Keras Tokenizer and fit on the sentences

tokenizer = Tokenizer();
tokenizer.fit_on_texts(sentences);
text_sequences = np.array(tokenizer.texts_to_sequences(sentences));
sequence_dict = tokenizer.word_index;
word_dict = dict((num, val) for (val, num) in sequence_dict.items());

# We get a map of encoding-to-word in sequence_dict

ravishchawla

'''
Clean each document by removing unnecesary characters and splitting by space.
'''
def clean_document(doco):
    punctuation = string.punctuation + '\n\n';
    punc_replace = ''.join([' ' for s in punctuation]);
    doco_clean = doco.replace('-', ' ');
    doco_alphas = re.sub(r'\W +', '', doco_clean)
    trans_table = str.maketrans(punctuation, punc_replace);
    doco_clean = ' '.join([word.translate(trans_table) for word in doco_alphas.split(' ')]);

ravishchawla

'''
Read reviews from a JSON-formatted file into an array.
'''
lines = [];
num_pos = 0; num_neg = 0; num_total = 75000;

with open('data/review.json', 'r') as f:
    for line in f:
        if (len(lines) >= (num_total * 2)):
            break;

ravishchawla

import time;
import datetime;
import requests;
import json;
import os;
import os.path;
import time;

'''
Bot to post Pull Request changes to a Slack channel.

ravishchawla

# coding: utf-8

# # Training a Word2Vec Model on the Reddit Comments Dataset
# 
# ### Ravish Chawla

# In[276]:

get_ipython().magic('matplotlib inline')