NLP for ChatBots : How to Train Simple Neural Networks for Sentiment Analysis

Data File: D:/mystuff/comments.txt

Data:

i love you,pos
i hate you,neg
i kill you,neg
i love to kill you,neg
i admire you,pos

import numpy as np

from sklearn.feature_extraction.text import TfidfVectorizer

def  tfidf(x):

   vec = TfidfVectorizer()

   vec.fit(x)

   return vec.transform(x).toarray() 

f = open('D:/mystuff/comments.txt')

lines = f.readlines()

text =[]

y=[]

for line in lines:

  w = line.lower().strip().split(",")

  text.append(w[0])

  l=0

  if w[1]=='pos':

    l=1

  y.append(l)

X = tfidf(text)

Y = np.c_[y]

print(X)

[[0.         0.         0.         0.861037   0.         0.50854232]
 [0.         0.90275015 0.         0.         0.         0.43016528]
 [0.         0.         0.861037   0.         0.         0.50854232]
 [0.         0.         0.50733821 0.50733821 0.62883263 0.29964212]
 [0.90275015 0.         0.         0.         0.         0.43016528]]

Print(Y)

[[1]
 [0]
 [0]
 [0]
 [1]]

def mse(y,ycap):
    return ((y-ycap)**2).mean()

def sigmoid(x):
    return 1/(1+np.exp(-x))

def derivative(x):
    return x*(1-x)

np.random.seed(101)

W1 = np.random.random((6,9))

W2 = np.random.random((9,1))

print(W1)

[[0.51639863 0.57066759 0.02847423 0.17152166 0.68527698 0.83389686
  0.30696622 0.89361308 0.72154386]
 [0.18993895 0.55422759 0.35213195 0.1818924  0.78560176 0.96548322
  0.23235366 0.08356143 0.60354842]
 [0.72899276 0.27623883 0.68530633 0.51786747 0.04848454 0.13786924
  0.18696743 0.9943179  0.5206654 ]
 [0.57878954 0.73481906 0.54196177 0.91315356 0.80792015 0.40299783
  0.35722434 0.95287671 0.34363158]
 [0.86509982 0.83027771 0.53816145 0.92246937 0.09714648 0.10284749
  0.7015073  0.89047987 0.1595603 ]
 [0.27557254 0.67249153 0.16430312 0.70137114 0.48763522 0.68067777
  0.52154819 0.04339669 0.2239366 ]]

print(W2)

[[0.57520509]
 [0.12043366]
 [0.50011671]
 [0.13800957]
 [0.0528084 ]
 [0.17827692]
 [0.44236813]
 [0.87758732]
 [0.94926413]]

pe=0
j=0
conv = 0.000000001
for i in range(100000):
    l1 = sigmoid(X.dot(W1))
    l2 = sigmoid(l1.dot(W2))
    e2 = Y-l2
    ce = mse(Y,l2)
    if i%50 ==0:
         print("Current Error ", ce)
    diff = abs(pe-ce)
    
    if diff<=conv:
          print("Training completed with ", i+1 , " iterations ")
          j=1  
          break
    d2 =  e2*derivative(l2)
    e1 = d2.dot(W2.T)
    d1 = e1*derivative(l1)
    W1 += X.T.dot(d1)
    W2 += l1.T.dot(d2)
    pe=ce
if j==0:
    print("Training not completed. ")

Current Error  1.2986065698996415e-05
Current Error  1.2929648731037894e-05
Current Error  1.2873704747295813e-05
Current Error  1.2818227898072753e-05
Current Error  1.2763212429079132e-05
Current Error  1.270865267950278e-05
Current Error  1.2654543080128585e-05
Current Error  1.2600878151500743e-05
Current Error  1.2547652502128916e-05
Current Error  1.249486082673763e-05
Current Error  1.2442497904559593e-05
Current Error  1.2390558597665709e-05
Current Error  1.2339037849334168e-05
Current Error  1.2287930682461896e-05
Current Error  1.2237232198009065e-05
Current Error  1.2186937573481663e-05
Training completed with  13264  iterations 

print(W1)

[[ 0.45603253 -0.62798423 -1.32964013 -1.43802362  0.49706007  0.54483843
   0.31845807  2.16208593  2.21925376]
 [ 0.22947915  1.38267337  1.56981474  1.49024135  0.89741654  1.15530848
   0.19881147 -1.48134687 -1.43274049]
 [ 0.77142345  1.44387767  2.23154546  2.02825123  0.26937912  0.52286594
   0.14791142 -1.36866899 -2.3683224 ]
 [ 0.47004479 -0.77242135 -1.26186218 -1.25233567  0.71075634  0.104229
   0.37518345  1.97398988  1.91398429]
 [ 0.91721091  1.62965816  1.76484005  1.98120147  0.17339261  0.34648761
   0.66978598 -0.82839166 -2.04960283]
 [ 0.20165153 -0.08539685 -0.63926595 -0.33355789  0.48797489  0.56822437
   0.5136963  -0.07133713  0.24117176]]

print(W2)

[[-0.21504125]
 [-2.45982184]
 [-3.83090743]
 [-3.7231175 ]
 [ 0.21219171]
 [-0.31003557]
 [ 0.1346712 ]
 [ 3.97735277]
 [ 5.25740624]]