Update app.py

app.py

@@ -1,6 +1,6 @@
 import gradio as gr
 from datasets import load_dataset
-from sentence_transformers import SentenceTransformer
+from sentence_transformers import SentenceTransformer, util
 import faiss
 import numpy as np
 from transformers import pipeline
@@ -51,10 +51,11 @@ def extract_each_feature(sample):
         "directions": directions,
     }
 
+
 # Apply preprocessing
 dataset = dataset.map(extract_each_feature)
 
-# --- 2. EMBEDDING AND RECOMMENDATION ENGINE (OPTIMIZED) ---
+# --- 2. EMBEDDING AND RECOMMENDATION ENGINE ---
 print("Loading embedding model...")
 model_name = "all-MiniLM-L6-v2"
 embedding_model = SentenceTransformer(f"sentence-transformers/{model_name}")
@@ -62,16 +63,15 @@ embedding_model = SentenceTransformer(f"sentence-transformers/{model_name}")
 index_file = "recipe_index.faiss"
 
 print(f"Loading FAISS index from {index_file}...")
-# Load the pre-computed FAISS index
 index = faiss.read_index(index_file)
 print(f"Index is ready. Total vectors in index: {index.ntotal}")
 
-# --- 3. SYNTHETIC GENERATION (IMPROVED MODEL AND PROMPT) ---
+# --- 3. SYNTHETIC GENERATION (IMPROVED) ---
 print("Loading generative model...")
 generator = pipeline('text-generation', model='gpt2-medium')
 
 def get_recommendations_and_generate(query_ingredients, k=3):
-    # --- 1. Get Recommendations ---
+    # 1. Get Recommendations
     query_vector = embedding_model.encode([query_ingredients])
     query_vector = np.array(query_vector, dtype=np.float32)
     distances, indices = index.search(query_vector, k)
@@ -86,77 +86,42 @@ def get_recommendations_and_generate(query_ingredients, k=3):
         }
         results.append(recipe)
         
-    # Defensive check: Ensure there are always 3 recommendations
     while len(results) < 3:
         results.append({"title": "No recipe found", "ingredients": "", "directions": ""})
-
-    # --- 2. Generate and Parse a New Recipe (with error handling) ---
-    generated_recipe = {
-        "title": "AI Recipe Generation Failed",
-        "ingredients": "The model could not generate a recipe for these ingredients.",
-        "directions": "Please try a different combination of ingredients."
-    }
-    
-    try:
-        prompt = f"Create a simple and delicious recipe using the following ingredients: {query_ingredients}."
-        
-        generated_outputs = generator(prompt, max_new_tokens=250, num_return_sequences=1, pad_token_id=50256)
-        
-        # Check if the model returned a valid output
-        if not generated_outputs or 'generated_text' not in generated_outputs[0]:
-            raise ValueError("Model did not return generated_text.")
-            
-        generated_text = generated_outputs[0]['generated_text'].replace(prompt, "").strip()
-        
-        # Parsing logic
-        lines = generated_text.split('\n')
-        if not lines or lines[0] == "":
-            raise ValueError("Generated text is empty.")
-            
-        title = lines[0].strip()
-        
-        ing_index = -1
-        dir_index = -1
-        for i, line in enumerate(lines):
-            if "ingredients" in line.lower() and ing_index == -1:
-                ing_index = i
-            if "directions" in line.lower() and dir_index == -1:
-                dir_index = i
         
-        if ing_index != -1 and dir_index != -1:
-            ingredients = "\n".join(lines[ing_index+1:dir_index]).strip()
-            directions = "\n".join(lines[dir_index+1:]).strip()
-        elif ing_index != -1:
-            ingredients = "\n".join(lines[ing_index+1:]).strip()
-            directions = "Not provided."
-        elif dir_index != -1:
-            directions = "\n".join(lines[dir_index+1:]).strip()
-            ingredients = "Not provided."
-        else:
-            ingredients = "Not provided."
-            directions = "\n".join(lines[1:]).strip() if len(lines) > 1 else ""
-
-        generated_recipe = {
-            "title": title,
-            "ingredients": ingredients,
-            "directions": directions
-        }
-
-    except Exception as e:
-        print(f"An error occurred in get_recommendations_and_generate: {e}")
-        # The generated_recipe dictionary is already set to a default error message
+    # 2. Generate 10 new recipe ideas
+    prompt = f"Create a simple and delicious recipe using the following ingredients: {query_ingredients}."
+    generated_outputs = generator(prompt, max_new_tokens=200, num_return_sequences=10, pad_token_id=50256)
+    
+    # 3. Find the best recipe out of the 10 generated
+    generated_texts = [output['generated_text'].replace(prompt, "").strip() for output in generated_outputs]
     
-    return results[0], results[1], results[2], generated_recipe
+    # Embed all 10 generated texts
+    generated_embeddings = embedding_model.encode(generated_texts)
+    
+    # Calculate cosine similarity between the user's query and each generated text
+    similarities = util.cos_sim(query_vector, generated_embeddings)
+    
+    # Find the index of the most similar generated recipe
+    best_recipe_index = np.argmax(similarities)
+    best_generated_recipe = generated_texts[best_recipe_index]
+
+    return results[0], results[1], results[2], best_generated_recipe
 
 # --- 4. GRADIO USER INTERFACE ---
 def format_recipe(recipe):
+    # Formats the recommended recipes with markdown
     if not recipe or not recipe['title']:
         return "### No recipe found."
     return f"### {recipe['title']}\n**Ingredients:**\n{recipe['ingredients']}\n\n**Directions:**\n{recipe['directions']}"
 
+def format_generated_recipe(recipe_text):
+    # Formats the AI-generated recipe as simple text
+    return recipe_text
+
 def recipe_wizard(ingredients):
-    rec1, rec2, rec3, gen_rec = get_recommendations_and_generate(ingredients)
-    return format_recipe(rec1), format_recipe(rec2), format_recipe(rec3), format_recipe(gen_rec)
+    rec1, rec2, rec3, gen_rec_text = get_recommendations_and_generate(ingredients)
+    return format_recipe(rec1), format_recipe(rec2), format_recipe(rec3), format_generated_recipe(gen_rec_text)
 
 end_time = time.time()
 print(f"Models and data loaded in {end_time - start_time:.2f} seconds.")