File size: 4,333 Bytes
bd6bce1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 | import joblib
import torch
import numpy as np
from transformers import BertTokenizer, BertModel
import os
import warnings
# Suppress warnings from XGBoost and other libraries for a cleaner output
warnings.filterwarnings('ignore', category=UserWarning)
warnings.filterwarnings('ignore', category=FutureWarning)
# --- Configuration ---
MODELS_DIR = 'saved_models'
SVM_MODEL_PATH = os.path.join(MODELS_DIR, 'bert_svm_model.joblib')
XGB_MODEL_PATH = os.path.join(MODELS_DIR, 'bert_xgboost_model.joblib')
LE_PATH = os.path.join(MODELS_DIR, 'label_encoder.joblib')
# --- Helper Function for BERT Embedding ---
def get_bert_embedding(text, model, tokenizer):
"""Generates a BERT embedding for a single piece of text."""
# Ensure the model is in evaluation mode
model.eval()
# Tokenize the text and create tensors
inputs = tokenizer(text, return_tensors='pt', truncation=True, padding=True, max_length=512)
# Get the model output without calculating gradients
with torch.no_grad():
outputs = model(**inputs)
# Use the mean of the last hidden state as the sentence embedding
embedding = outputs.last_hidden_state.mean(dim=1).squeeze().numpy()
# Reshape for a single prediction (required by sklearn models)
return embedding.reshape(1, -1)
# --- Main Application ---
def run_intent_recognizer():
"""
Loads all necessary components and runs the interactive prediction loop.
"""
# 1. Load all required files
print("Loading models and encoders... This may take a moment.")
try:
model_svm = joblib.load(SVM_MODEL_PATH)
model_xgb = joblib.load(XGB_MODEL_PATH)
label_encoder = joblib.load(LE_PATH)
# Load BERT components
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
bert_model = BertModel.from_pretrained('bert-base-uncased')
print("All models loaded successfully! 🚀")
except FileNotFoundError as e:
print(f"\nError: A required file was not found: {e}")
print("Please make sure you have run all the necessary training scripts and that the following files exist:")
print(f"- {SVM_MODEL_PATH}")
print(f"- {XGB_MODEL_PATH}")
print(f"- {LE_PATH}")
return
# 2. Start the interactive prediction loop
print("\nEnter a sentence to recognize its intent. Type 'exit' to close.")
while True:
sentence = input("\nYour sentence: ")
if sentence.lower() == 'exit':
print("Goodbye! 👋")
break
if not sentence.strip():
print("Please enter a valid sentence.")
continue
# Generate embedding for the input sentence
embedding = get_bert_embedding(sentence, bert_model, tokenizer)
# --- SVM Prediction ---
pred_svm_idx = model_svm.predict(embedding)[0]
pred_svm_label = label_encoder.inverse_transform([pred_svm_idx])[0]
# Get confidence scores (probabilities)
# For SVM, we need to enable probability=True during training for this,
# but we can use decision_function as a proxy for confidence.
# For simplicity, we'll just show the prediction. If probabilities are needed,
# the SVM model needs to be retrained with probability=True.
# For now, we get probabilities if available.
try:
confidence_svm = np.max(model_svm.predict_proba(embedding)) * 100
print(f"\n🤖 SVM Model Prediction:")
print(f" Intent: {pred_svm_label}")
print(f" Confidence: {confidence_svm:.2f}%")
except AttributeError:
print(f"\n🤖 SVM Model Prediction:")
print(f" Intent: {pred_svm_label}")
print(f" (Confidence scores not available for this SVM model)")
# --- XGBoost Prediction ---
pred_xgb_idx = model_xgb.predict(embedding)[0]
pred_xgb_label = label_encoder.inverse_transform([pred_xgb_idx])[0]
confidence_xgb = np.max(model_xgb.predict_proba(embedding)) * 100
print(f"\n🚀 XGBoost Model Prediction:")
print(f" Intent: {pred_xgb_label}")
print(f" Confidence: {confidence_xgb:.2f}%")
print("-" * 30)
if __name__ == '__main__':
run_intent_recognizer()
|