zimspeak / app.py
Kumbirai's picture
Update app.py
fc93d1d verified
Raw History Blame Contribute Delete
3.85 kB
import gradio as gr
import tensorflow as tf
import numpy as np
import random
import json
import pickle
import nltk
# =============================
# 1) DOWNLOAD NLTK DATA FIX
# =============================
# Avoid "punkt_tab not found" errors on Hugging Face Spaces or other hosts
nltk.download("punkt_tab", quiet=True)
nltk.download("punkt", quiet=True)
nltk.download("wordnet", quiet=True)
nltk.download("omw-1.4", quiet=True)
lemmatizer = nltk.stem.WordNetLemmatizer()
# =============================
# 2) LOAD YOUR MODEL & FILES
# =============================
# Load your saved model (.h5)
model = tf.keras.models.load_model("chatbot_model.h5")
# Load words and classes from the .pkl files
with open("words.pkl", "rb") as f:
words = pickle.load(f)
with open("classes.pkl", "rb") as f:
classes = pickle.load(f)
# Load intents
with open("intents.json", "r") as f:
intents_data = json.load(f)
# For ignoring punctuation, etc.
ignore_chars = ['?', '!', '.', ',']
# =============================
# 3) HELPER FUNCTIONS
# =============================
def clean_up_sentence(sentence):
"""
Tokenize & lemmatize the user input.
"""
sentence_words = nltk.word_tokenize(sentence)
sentence_words = [
lemmatizer.lemmatize(word.lower())
for word in sentence_words
if word not in ignore_chars
]
return sentence_words
def bag_of_words(sentence, vocabulary):
"""
Convert the input sentence to a bag-of-words vector
matching the training vocabulary.
"""
sentence_words = clean_up_sentence(sentence)
bag = [0] * len(vocabulary)
for sw in sentence_words:
for i, w in enumerate(vocabulary):
if w == sw:
bag[i] = 1
return np.array(bag)
def predict_class(sentence):
"""
Predict the intent class for a given sentence.
Returns a list of dicts: [{intent, probability}, ...]
"""
bow = bag_of_words(sentence, words)
res = model.predict(np.array([bow]))[0]
ERROR_THRESHOLD = 0.25
results = []
for i, prob in enumerate(res):
if prob > ERROR_THRESHOLD:
results.append({"intent": classes[i], "probability": float(prob)})
results.sort(key=lambda x: x["probability"], reverse=True)
return results
def get_response(intents_list, intents_json):
"""
Pick a response from the top predicted intent
or return a default message if no predictions.
"""
if not intents_list:
return "I’m not sure I understand. Could you rephrase?"
top_intent = intents_list[0]["intent"]
for i in intents_json["intents"]:
if i["tag"] == top_intent:
return random.choice(i["responses"])
return "I’m not sure I understand. Could you try again?"
# =============================
# 4) GRADIO FUNCTION
# =============================
def chatbot_respond(user_input, history):
"""
Called by Gradio for each new message.
'history' is a list of (user_msg, bot_msg) so far (not used here).
"""
# Predict intent
intents = predict_class(user_input)
# Get response from the top intent
response = get_response(intents, intents_data)
return response
# =============================
# 5) CREATE GRADIO CHAT
# =============================
demo = gr.ChatInterface(
fn=chatbot_respond,
title="ZimSpeak: Emotional Support Bot",
description="""
**Disclaimer**:
I am an **AI-based chatbot**, not a licensed mental health professional.
I can offer empathy and general suggestions, **but I cannot provide professional advice**.
If you are in crisis or need immediate support, please reach out to a **licensed mental health professional** or **emergency services**.
""",
)
# =============================
# 6) LAUNCH (LOCAL TEST)
# =============================
if __name__ == "__main__":
demo.launch()