from __future__ import division

from math import *

import nltk

from nltk.metrics import *

import json

import zmq

from sentiment_analyzer import date_convert

from sentiment_analyzer import month_to_num

from datetime import datetime

import time, calendar

# The use of the term 'trigram' in this file is arbritrary, and can represent either unigram, bigram, or trigram functionality.

# Currently this classifier uses bigrams.

jsonFile = open("master_tweet_sample.json")

jsonString = jsonFile.read()

tweetRawData = json.loads(jsonString)

jsonFile.close()

tweetCounter = 0

testList = []

trainList = []

testSentimentDict = dict()

testTextDict = dict()

allTrigrams = ""

posTrigrams = ""

neuTrigrams = ""

negTrigrams = ""

for _dict in tweetRawData:

if (tweetCounter < 0): #setting to zero so that all tweets will be placed in the training CFD

testSentimentDict[_dict['id']] = _dict['sentiment']

testTextDict[_dict['id']] = _dict['text']

# ^^^ unigram

#textArray = _dict['text'].split()

#bigrams = ""

#for i in range(len(textArray)):

# bigram = ""

# if (i > 0):

# bigram = textArray[i-1] + textArray[i]

#bigrams += bigram + " "

#testTextDict[_dict['id']] = bigrams

# ^^^ bigram

testList.append(_dict['id'])

tweetCounter += 1

else:

trainList.append(_dict['id'])

textArray = _dict['text'].split()

trigrams = ""

for i in range(len(textArray)):

#trigram = ""

#if (i > 1):

#trigram = textArray[i-2] + textArray[i-1] + textArray[i]

#trigrams += trigram + " "

# ^^^ trigram

#trigram = ""

#if (i > 0):

# trigram = textArray[i-1] + textArray[i]

#trigrams += trigram + " "

# ^^^ bigram

trigram = textArray[i]

trigrams += trigram + " "

# ^^^ unigram

tweetCounter += 1

sentiment = _dict['sentiment']

if (sentiment == "positive"):

posTrigrams += trigrams

elif (sentiment == "negative"):

negTrigrams += trigrams

else:

neuTrigrams += trigrams

posTrigramListX = []

negTrigramListX = []

neuTrigramListX = []

posTrigramsX = ""

negTrigramsX = ""

neuTrigramsX = ""

posTrigramList = posTrigrams.split()

negTrigramList = negTrigrams.split()

neuTrigramList = neuTrigrams.split()

#Calculate ratios of classifications in raw training data

posRatio = len(posTrigramList)/len(allTrigramList)

negRatio = len(negTrigramList)/len(allTrigramList)

neuRatio = len(neuTrigramList)/len(allTrigramList)

# Using naive 'Information Gain' technique, to improve quality of CFD by removing N-grams of similiar frequency

discardThreshold = 300;

for trigram in posTrigramList:

calc1 = (posTrigramList.count(trigram) * posRatio) / ((neuTrigramList.count(trigram) * neuRatio) + 0.0001)

calc2 = (posTrigramList.count(trigram) * posRatio) / ((negTrigramList.count(trigram) * negRatio) + 0.0001)

if ( (discardThreshold < calc1 or calc1 < (1/discardThreshold)) or (discardThreshold < calc2 or calc2 < (1/discardThreshold))): posTrigramListX.append(trigram)

for trigram in negTrigramList:

calc1 = ((negTrigramList.count(trigram) * negRatio) + 0.0001) / ((posTrigramList.count(trigram) * posRatio) + 0.0001)

calc2 = ((negTrigramList.count(trigram) * negRatio) + 0.0001)/ ((neuTrigramList.count(trigram) * neuRatio) + 0.0001)

if ( (discardThreshold < calc1 or calc1 < (1/discardThreshold)) or (discardThreshold < calc2 or calc2 < (1/discardThreshold))): negTrigramListX.append(trigram)

for trigram in neuTrigramList:

calc1 = ((neuTrigramList.count(trigram) * neuRatio) + 0.0001) / ((negTrigramList.count(trigram) * negRatio) + 0.0001)

calc2 = ((neuTrigramList.count(trigram) * neuRatio) + 0.0001) / ((posTrigramList.count(trigram) * posRatio) + 0.0001)

if ( (discardThreshold < calc1 or calc1 < (1/discardThreshold)) or (discardThreshold < calc2 or calc2 < (1/discardThreshold))): neuTrigramListX.append(trigram)

for trigram in posTrigramListX:

posTrigramsX += " " + trigram

for trigram in negTrigramListX:

negTrigramsX += " " + trigram

for trigram in neuTrigramListX:

neuTrigramsX += " " + trigram

trainDictX = dict()

trainDictX['positive'] = posTrigramsX

trainDictX['negative'] = negTrigramsX

trainDictX['neutral'] = neuTrigramsX

trainCFD = nltk.ConditionalFreqDist(

(sentiment, trigram)

for sentiment in trainDictX.keys()

for trigram in trainDictX[sentiment].split())

#actualScoreList = []

#testScoreList = []

#Print statements to analyze contents of the training CFD

#print "Analyzing %d total tri-grams." % trainCFD.N()

#print "There are %d postively rated trigrams." % trainCFD['positive'].N()

#print "There are %d negatively rated trigrams." % trainCFD['negative'].N()

#print "There are %d neutrally rated trigrams." % trainCFD['neutral'].N()

#print trainCFD.keys()

# Create a port for recieving data on port 5556 (for get_tweets.py)

contextIN = zmq.Context()

socketIN = contextIN.socket(zmq.REP)

socketIN.bind("tcp://*:5556")

# Connect to the zmq server and prepare it to send data

contextOUT = zmq.Context()

socketOUT = contextOUT.socket(zmq.REQ)

socketOUT.connect("tcp://localhost:5555")

# Begin processing tweets from the get_tweets client.

while True:

print "Waiting to recieve a message"

messageIN = socketIN.recv()

rcvd = json.loads(messageIN)

print "Message accepted, processing"

for tweet in rcvd:

# Handler for tweet_send type.

if tweet['type'] == "tweet_send":

date = date_convert(tweet)

best = -9999

win = 'none'

for rating in trainCFD.keys():

logProb = 0

for word in tweet['text'].split():

logProb += log( ((trainCFD[rating][word] + 0.001)/(trainCFD[rating].N() + 0.001)) * (trainCFD[rating].N()/trainCFD.N()))

if (logProb > best):

best = logProb

win = rating

# Formatted JSON that will be sent to the server.

data_set = {'type': "tweet_push", 'company':tweet["company"], 'date': date, 'sentiment' : win, 'id' : tweet["id"],'tweet' : tweet['text']}

# Send data to the zmq server

messageOUT = json.dumps(data_set)

socketOUT.send(messageOUT)

messageOUT = socketOUT.recv()

# Shutdown protocol.

elif rcvd['type'] == "tweet_stop":

print "Recieved tweet_stop."

print "Shutting down..."

socketIN.send("Ack")

sys.exit()

else:

# Send reply back to client that the query is unspecified.

print "received unknown query, ignoring"

socketIN.send("Ack")

socketIN.send("Ack")

#Analzying results for testing the Classifier

#cm = ConfusionMatrix(testScoreList, actualScoreList)

#print cm

#score = 0

#for i in range(len(actualScoreList)):

# if (testScoreList[i] == actualScoreList[i]):

# score += 1

#overallScore = (score/len(testScoreList)) * 100

#print "Overall accuracy is ", overallScore, "%."