Sentimentron

def build_model(tparams, options, maxw, training=True):
    # Input variables
    xc0 = tensor.tensor3('xc0', dtype='int8')
    xc1 = tensor.tensor3('xc1', dtype='int8')
    mask0 = tensor.tensor4('mask0', dtype=config.floatX)
    mask1 = tensor.tensor4('mask1', dtype=config.floatX)
    y = tensor.vector('y', dtype='int8')
    y_mask = tensor.vector('y_mask', dtype='float32')

    n_batch = xc0.shape[2]

Sentimentron

import csv
import numpy
import scipy
from sklearn.linear_model import SGDClassifier
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.naive_bayes import MultinomialNB

def train_test_split(x, y, test_size=0.05):
    l = int(len(x) * test_size)

Sentimentron

var sqlite3 = require('sqlite3').verbose();
var fileExists = require('file-exists');
var dracula = require('dracula-sentiment');
var request = require('request');
// This is the database file written to by the collection script...
var databaseFile = "tweets.sqlite";
var db = new sqlite3.Database(databaseFile);
// ...and this is the database we'll generate.
var outDatabaseFile = "analysis.sqlite"
var createDatabase = !fileExists(outDatabaseFile);

Sentimentron

#!/bin/bash

# This is the “Consumer Key (API Key)” field on apps.twitter.com
export TWITTER_CONSUMER_KEY=""
# The is the “Consumer Secret (API Secret)” field on apps.twitter.com
export TWITTER_CONSUMER_SECRET=""
# This is the “Access Token” from the “Your Access Token” section
export TWITTER_ACCESS_TOKEN_KEY=""
# This is the “Access Token Secret” from the same section
export TWITTER_ACCESS_TOKEN_SECRET=""

Sentimentron

// collect.js - Track a Twitter topic and dump it into a SQLite database
var sqlite3 = require('sqlite3').verbose();
var fileExists = require('file-exists');
var request = require('request');
var Twitter = require("twitter");
_ = require('lodash')

// If the output database doesn't exist, create it.
var databaseFile = "tweets.sqlite";
var createDatabase = !fileExists(databaseFile);

Sentimentron

with tf.Graph().as_default():
  tparams = init_tparams(params)
  # Initialize the TensorFlow session
  sm = tf.train.SessionManager(ready_op=tf.assert_variables_initialized())
  # Create TensorFlow variables from the parameters
  saver = tf.train.Saver(tparams)
  # ...
  sess = sm.prepare_session("", init_op=tf.initialize_all_variables(),
                                saver=saver)
  # ... after completing a certain number of minibatches ...

Sentimentron

    def cost_fn(cur):
        print options['ydim']
        # TensorFlow does not support negative indexing WTF
        # labels = cur[:, :, -1]
        labels = cur[:, options['ydim']] # TODO: don't hard-code me
        labels = tf.cast(labels, dtype='int32')
        preds = cur[:, :options['ydim']]
        # The non-sparse version of this: y
        return tf.nn.sparse_softmax_cross_entropy_with_logits(preds, labels)
        

Sentimentron

def cost_scan_i(i, j, free_var):
    return -tensor.log(i[tensor.arange(n_batch), j] + 1e-8)
cost, _ = theano.scan(cost_scan_i, outputs_info=None, sequences=[pred, y, tensor.arange(n_batch)])
cost = cost.mean()

Sentimentron

def lstm_layer(tparams, state_below, options, prefix='lstm', mask=None, go_backwards=False, mult=1):
    nsteps = state_below.shape[0]
    if state_below.ndim == 3:
        n_samples = state_below.shape[1]
    else:
        n_samples = 1

    assert mask is not None

    def _slice(_x, n, dim):

Sentimentron

def lstm_bidirectional_layer(tparams, state_below, options, prefix='lstm'):
    def _p(pp, name):
        return '%s_%s' % (pp, name)

    lstm_fw_cell = rnn_cell.BasicLSTMCell(32, forget_bias=1.0)
    lstm_bw_cell = rnn_cell.BasicLSTMCell(32, forget_bias=1.0) # TODO: don't hard-code me

    lstm_fw_multicell = rnn_cell.MultiRNNCell([lstm_fw_cell]*options['letter_layers'])
    lstm_bw_multicell = rnn_cell.MultiRNNCell([lstm_bw_cell]*options['letter_layers'])