# conda install -c conda-forge textblob

import pandas as pd
import numpy as np
from sklearn.feature_extraction.text import CountVectorizer,TfidfTransformer

# Loading the data set

email_data = pd.read_csv("~/Downloads/Data Science/data set/ham_spam.csv",encoding = "ISO-8859-1")


# cleaning data 
import re
stop_words = []
with open("D:\\Assignment\\words_data\\stop.txt") as f:
    stop_words = f.read()


# splitting the entire string by giving separator as "\n" to get list of 
# all stop words
stop_words = stop_words.split("\n")


"this is awsome 1231312 $#%$# a i he yu nwj"

def cleaning_text(i):
    i = re.sub("[^A-Za-z" "]+"," ",i).lower()
    i = re.sub("[0-9" "]+"," ",i)
    w = []
    for word in i.split(" "):
        if len(word)>3:
            w.append(word)
    return (" ".join(w))

"This is Awsome 1231312 $#%$# a i he yu nwj".split(" ")

cleaning_text("This is Awsome 1231312 $#%$# a i he yu nwj")


              
# testing above function with sample text => removes punctuations, numbers
cleaning_text("Hope you are having a good week. Just checking in")
cleaning_text("hope i can understand your feelings 123121. 123 hi how .. are you?")

email_data.text = email_data.text.apply(cleaning_text)

# removing empty rows 
email_data.shape
email_data = email_data.loc[email_data.text != " ",:]


# CountVectorizer
# Convert a collection of text documents to a matrix of token counts

# TfidfTransformer
# Transform a count matrix to a normalized tf or tf-idf representation

# creating a matrix of token counts for the entire text document 

def split_into_words(i):
    return [word for word in i.split(" ")]


# splitting data into train and test data sets 
from sklearn.model_selection import train_test_split

email_train,email_test = train_test_split(email_data,test_size=0.3)


# Preparing email texts into word count matrix format 
emails_bow = CountVectorizer(analyzer=split_into_words).fit(email_data.text)

# ["mailing","body","texting"]
# ["mailing","awesome","good"]

# ["mailing","body","texting","good","awesome"]



#        "mailing" "body" "texting" "good" "awesome"
#  0          1        1       1        0       0
 
#  1          1        0        0       1       1    

# For all messages
all_emails_matrix = emails_bow.transform(email_data.text)
all_emails_matrix.shape # (5559,6661)
# For training messages
train_emails_matrix = emails_bow.transform(email_train.text)
train_emails_matrix.shape # (3891,6661)



# For testing messages
test_emails_matrix = emails_bow.transform(email_test.text)
test_emails_matrix.shape # (1668,6661)

####### Without TFIDF matrices ########################
# Preparing a naive bayes model on training data set 

from sklearn.naive_bayes import MultinomialNB as MB
from sklearn.naive_bayes import GaussianNB as GB

# Multinomial Naive Bayes
classifier_mb = MB()
classifier_mb.fit(train_emails_matrix,email_train.type)
train_pred_m = classifier_mb.predict(train_emails_matrix)
accuracy_train_m = np.mean(train_pred_m==email_train.type) # 98%

test_pred_m = classifier_mb.predict(test_emails_matrix)
accuracy_test_m = np.mean(test_pred_m==email_test.type) # 96%

# Gaussian Naive Bayes 
classifier_gb = GB()
classifier_gb.fit(train_emails_matrix.toarray(),email_train.type.values) # we need to convert tfidf into array format which is compatible for gaussian naive bayes
train_pred_g = classifier_gb.predict(train_emails_matrix.toarray())
accuracy_train_g = np.mean(train_pred_g==email_train.type) # 90%

test_pred_g = classifier_gb.predict(test_emails_matrix.toarray())
accuracy_test_g = np.mean(test_pred_g==email_test.type) # 83%


#########################################################3

# Learning Term weighting and normalizing on entire emails
tfidf_transformer = TfidfTransformer().fit(all_emails_matrix)

# Preparing TFIDF for train emails
train_tfidf = tfidf_transformer.transform(train_emails_matrix)

train_tfidf.shape # (3891, 6661)

# Preparing TFIDF for test emails
test_tfidf = tfidf_transformer.transform(test_emails_matrix)

test_tfidf.shape #  (1668, 6661)

# Preparing a naive bayes model on training data set 

from sklearn.naive_bayes import MultinomialNB as MB
from sklearn.naive_bayes import GaussianNB as GB

# Multinomial Naive Bayes
classifier_mb = MB()
classifier_mb.fit(train_tfidf,email_train.type)
train_pred_m = classifier_mb.predict(train_tfidf)
accuracy_train_m = np.mean(train_pred_m==email_train.type) # 96%

test_pred_m = classifier_mb.predict(test_tfidf)
accuracy_test_m = np.mean(test_pred_m==email_test.type) # 95%

# Gaussian Naive Bayes 
classifier_gb = GB()
classifier_gb.fit(train_tfidf.toarray(),email_train.type.values) # we need to convert tfidf into array format which is compatible for gaussian naive bayes
train_pred_g = classifier_gb.predict(train_tfidf.toarray())
accuracy_train_g = np.mean(train_pred_g==email_train.type) # 91%
test_pred_g = classifier_gb.predict(test_tfidf.toarray())
accuracy_test_g = np.mean(test_pred_g==email_test.type) # 85%

# inplace of tfidf we can also use train_emails_matrix and test_emails_matrix instead of term inverse document frequency matrix