Spaces:

Yhhxhfh
/

Ghgg

Building

App Files Files Community

Yhhxhfh commited on 1 day ago

Commit

6895902

•

1 Parent(s): 40c33c8

Update app.py

Browse files

Files changed (1) hide show

app.py +17 -17

app.py CHANGED Viewed

@@ -2,8 +2,8 @@ import uvicorn
 import nltk
 nltk.download('punkt')
 nltk.download('wordnet')
-nltk.download('omw-1.4')
 nltk.download('punkt_tab')
 nltk.download('averaged_perceptron_tagger')
 from nltk.stem import WordNetLemmatizer
 from nltk.corpus import wordnet
@@ -16,7 +16,6 @@ import random
 import asyncio
 import concurrent.futures
 import multiprocessing
-import io
 import numpy as np
 from tensorflow.keras import Sequential
@@ -57,11 +56,16 @@ async def train_and_save_model():
         documents = []
         ignore_words = ['?', '!']
-        if not r.exists('intents'):
             intents = {"intents": []}
-            r.set('intents', json.dumps(intents))
-        else:
-            intents = json.loads(r.get('intents'))
         print("Loading user questions from Redis...")
         if not r.exists('user_questions_loaded'):
@@ -79,7 +83,7 @@ async def train_and_save_model():
                         classes.append("unknown")
             r.set('user_questions_loaded', 1)
-        print("Processing intents from Redis...")
         for intent in intents['intents']:
             for pattern in intent['patterns']:
                 w = nltk.word_tokenize(pattern)
@@ -130,9 +134,8 @@ async def train_and_save_model():
         train_y = np.array([row[1] for row in training])
         print("Loading or creating model...")
-        if r.exists('chatbot_model'):
-            with io.BytesIO(r.get('chatbot_model')) as f:
-                model = load_model(f)
         else:
             input_layer = Input(shape=(len(train_x[0]),))
             layer1 = Dense(128, activation='relu')(input_layer)
@@ -147,15 +150,13 @@ async def train_and_save_model():
             model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
         print("Training the model...")
-        model.fit(train_x, train_y, epochs=1, batch_size=len(train_x), verbose=0, callbacks=[TqdmCallback(verbose=2)])
         print("Saving data to Redis...")
         r.set('words', pickle.dumps(words))
         r.set('classes', pickle.dumps(classes))
-        with io.BytesIO() as f:
-            save_model(model, f)
-            r.set('chatbot_model', f.getvalue())
         print("Data and model saved. Re-training...")
@@ -182,8 +183,7 @@ class ChatMessage(BaseModel):
 async def chat(message: ChatMessage):
     words = pickle.loads(r.get('words'))
     classes = pickle.loads(r.get('classes'))
-    with io.BytesIO(r.get('chatbot_model')) as f:
-        model = load_model(f)
     sentence_words = nltk.word_tokenize(message.message)
     sentence_words = [lemmatizer.lemmatize(word.lower()) for word in sentence_words]
@@ -211,7 +211,7 @@ async def tag_question(question: str, tag: str):
     r.set(f"tag:{question}", tag)
     return {"message": "Tag saved"}
-html_code = """
 <!DOCTYPE html>
 <html>
 <head>

 import nltk
 nltk.download('punkt')
 nltk.download('wordnet')
 nltk.download('punkt_tab')
+nltk.download('omw-1.4')
 nltk.download('averaged_perceptron_tagger')
 from nltk.stem import WordNetLemmatizer
 from nltk.corpus import wordnet
 import asyncio
 import concurrent.futures
 import multiprocessing
 import numpy as np
 from tensorflow.keras import Sequential
         documents = []
         ignore_words = ['?', '!']
+        try:
+            with open('intents.json') as file:
+                intents = json.load(file)
+        except FileNotFoundError:
             intents = {"intents": []}
+            with open('intents.json', 'w') as file:
+                json.dump(intents, file, indent=4)
+            print("intents.json created. Please populate it with training data.")
+            await asyncio.sleep(60)
+            continue
         print("Loading user questions from Redis...")
         if not r.exists('user_questions_loaded'):
                         classes.append("unknown")
             r.set('user_questions_loaded', 1)
+        print("Processing intents from intents.json...")
         for intent in intents['intents']:
             for pattern in intent['patterns']:
                 w = nltk.word_tokenize(pattern)
         train_y = np.array([row[1] for row in training])
         print("Loading or creating model...")
+        if os.path.exists('chatbot_model.h5'):
+            model = load_model('chatbot_model.h5')
         else:
             input_layer = Input(shape=(len(train_x[0]),))
             layer1 = Dense(128, activation='relu')(input_layer)
             model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
         print("Training the model...")
+        model.fit(train_x, train_y, epochs=1, batch_size=len(train_x), verbose=0, callbacks=[TqdmCallback(verbose=2)])
         print("Saving data to Redis...")
         r.set('words', pickle.dumps(words))
         r.set('classes', pickle.dumps(classes))
+        save_model(model, 'chatbot_model.h5')
         print("Data and model saved. Re-training...")
 async def chat(message: ChatMessage):
     words = pickle.loads(r.get('words'))
     classes = pickle.loads(r.get('classes'))
+    model = load_model('chatbot_model.h5')
     sentence_words = nltk.word_tokenize(message.message)
     sentence_words = [lemmatizer.lemmatize(word.lower()) for word in sentence_words]
     r.set(f"tag:{question}", tag)
     return {"message": "Tag saved"}
+html_code = """
 <!DOCTYPE html>
 <html>
 <head>