Upload training.py
Browse files- training.py +148 -0
training.py
ADDED
@@ -0,0 +1,148 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch.utils.data import Dataset,DataLoader
|
3 |
+
import torch.nn as nn
|
4 |
+
import nltk
|
5 |
+
from nltk.stem.porter import PorterStemmer
|
6 |
+
import json
|
7 |
+
import numpy as np
|
8 |
+
|
9 |
+
|
10 |
+
def Training():
|
11 |
+
|
12 |
+
class NeuralNet(nn.Module):
|
13 |
+
|
14 |
+
def __init__(self,input_size,hidden_size,num_classes):
|
15 |
+
super(NeuralNet,self).__init__()
|
16 |
+
self.l1 = nn.Linear(input_size,hidden_size)
|
17 |
+
self.l2 = nn.Linear(hidden_size,hidden_size)
|
18 |
+
self.l3 = nn.Linear(hidden_size,num_classes)
|
19 |
+
self.relu = nn.ReLU()
|
20 |
+
|
21 |
+
def forward(self,x):
|
22 |
+
out = self.l1(x)
|
23 |
+
out = self.relu(out)
|
24 |
+
out = self.l2(out)
|
25 |
+
out = self.relu(out)
|
26 |
+
out = self.l3(out)
|
27 |
+
return out
|
28 |
+
|
29 |
+
Stemmer = PorterStemmer()
|
30 |
+
|
31 |
+
def tokenize(sentence):
|
32 |
+
return nltk.word_tokenize(sentence)
|
33 |
+
|
34 |
+
def stem(word):
|
35 |
+
return Stemmer.stem(word.lower())
|
36 |
+
|
37 |
+
def bag_of_words(tokenized_sentence,words):
|
38 |
+
sentence_word = [stem(word) for word in tokenized_sentence]
|
39 |
+
bag = np.zeros(len(words),dtype=np.float32)
|
40 |
+
|
41 |
+
for idx , w in enumerate(words):
|
42 |
+
if w in sentence_word:
|
43 |
+
bag[idx] = 1
|
44 |
+
|
45 |
+
return bag
|
46 |
+
|
47 |
+
with open("intents.json",'r') as f:
|
48 |
+
intents = json.load(f)
|
49 |
+
|
50 |
+
all_words = []
|
51 |
+
tags = []
|
52 |
+
xy = []
|
53 |
+
|
54 |
+
for intent in intents['intents']:
|
55 |
+
tag = intent['tag']
|
56 |
+
tags.append(tag)
|
57 |
+
|
58 |
+
for pattern in intent['patterns']:
|
59 |
+
w = tokenize(pattern)
|
60 |
+
all_words.extend(w)
|
61 |
+
xy.append((w,tag))
|
62 |
+
|
63 |
+
ignore_words = [',','?','/','.','!']
|
64 |
+
all_words = [stem(w) for w in all_words if w not in ignore_words]
|
65 |
+
all_words = sorted(set(all_words))
|
66 |
+
tags = sorted(set(tags))
|
67 |
+
|
68 |
+
x_train = []
|
69 |
+
y_train = []
|
70 |
+
|
71 |
+
for (pattern_sentence,tag) in xy:
|
72 |
+
bag = bag_of_words(pattern_sentence,all_words)
|
73 |
+
x_train.append(bag)
|
74 |
+
|
75 |
+
label = tags.index(tag)
|
76 |
+
y_train.append(label)
|
77 |
+
|
78 |
+
x_train = np.array(x_train)
|
79 |
+
y_train = np.array(y_train)
|
80 |
+
|
81 |
+
num_epochs = 1000
|
82 |
+
batch_size = 8
|
83 |
+
learning_rate = 0.001
|
84 |
+
input_size = len(x_train[0])
|
85 |
+
hidden_size = 8
|
86 |
+
output_size = len(tags)
|
87 |
+
print(">> Training The Chats Module :- Conciousness ")
|
88 |
+
|
89 |
+
class ChatDataset(Dataset):
|
90 |
+
|
91 |
+
def __init__(self):
|
92 |
+
self.n_samples = len(x_train)
|
93 |
+
self.x_data = x_train
|
94 |
+
self.y_data = y_train
|
95 |
+
|
96 |
+
def __getitem__(self,index):
|
97 |
+
return self.x_data[index],self.y_data[index]
|
98 |
+
|
99 |
+
def __len__(self):
|
100 |
+
return self.n_samples
|
101 |
+
|
102 |
+
dataset = ChatDataset()
|
103 |
+
|
104 |
+
train_loader = DataLoader(dataset=dataset,
|
105 |
+
batch_size=batch_size,
|
106 |
+
shuffle=True,
|
107 |
+
num_workers=0)
|
108 |
+
|
109 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
110 |
+
model = NeuralNet(input_size,hidden_size,output_size).to(device=device)
|
111 |
+
criterion = nn.CrossEntropyLoss()
|
112 |
+
optimizer = torch.optim.Adam(model.parameters(),lr=learning_rate)
|
113 |
+
|
114 |
+
for epoch in range(num_epochs):
|
115 |
+
for (words,labels) in train_loader:
|
116 |
+
words = words.to(device)
|
117 |
+
labels = labels.to(dtype=torch.long).to(device)
|
118 |
+
outputs = model(words)
|
119 |
+
loss = criterion(outputs,labels)
|
120 |
+
optimizer.zero_grad()
|
121 |
+
loss.backward()
|
122 |
+
optimizer.step()
|
123 |
+
|
124 |
+
if (epoch+1) % 100 ==0:
|
125 |
+
print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {loss.item():.4f}')
|
126 |
+
|
127 |
+
print(f'Final Loss : {loss.item():.4f}')
|
128 |
+
|
129 |
+
data = {
|
130 |
+
"model_state":model.state_dict(),
|
131 |
+
"input_size":input_size,
|
132 |
+
"hidden_size":hidden_size,
|
133 |
+
"output_size":output_size,
|
134 |
+
"all_words":all_words,
|
135 |
+
"tags":tags
|
136 |
+
}
|
137 |
+
|
138 |
+
FILE = "intents.pth"
|
139 |
+
torch.save(data,FILE)
|
140 |
+
|
141 |
+
print(f"Training Complete, File Saved To {FILE}")
|
142 |
+
print(" ")
|
143 |
+
|
144 |
+
|
145 |
+
|
146 |
+
|
147 |
+
Training()
|
148 |
+
|