Create app.py

app.py

@@ -0,0 +1,104 @@
+import pandas as pd
+import numpy as np
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.metrics.pairwise import cosine_similarity
+from transformers import pipeline, HfApi, HfFolder
+import gradio as gr
+
+# Hugging Face login - Ensure that you are logged into Hugging Face before running this space
+HfFolder.save_token("<your_hugging_face_token>")  # This is set once to avoid token exposure in the code
+
+# Load and preprocess the data
+def preprocess_data(file_path):
+    """Load and preprocess the CSV data."""
+    data = pd.read_csv(file_path)
+
+    # Clean column names
+    data.columns = data.columns.str.strip().str.replace('#', 'Count').str.replace(' ', '_')
+
+    # Handle missing values (if any)
+    data.fillna(0, inplace=True)
+
+    return data
+
+# Convert data into a retrievable knowledge base
+def create_knowledge_base(data):
+    """Transform the data into a knowledge base suitable for retrieval."""
+    # Combine relevant fields into a single text-based feature for embedding
+    data['Knowledge_Text'] = data.apply(lambda row: (
+        f"Player: {row['Player_Name']}, Position: {row['Main_Possition']}, "
+        f"Date: {row['Date']}, Session: {row['Session_Name']}, "
+        f"Played Time: {row['Played_Time_(min)']} minutes, Top Speed: {row['Top_Speed_(km/h)']} km/h, "
+        f"Distance Covered: {row['Dist._Covered_(m)']} meters, "
+        f"Intensity: {row['Session_Intensity']}, "
+        f"RPE: {row['RPE']}, s-RPE: {row['s-RPE']}"
+    ), axis=1)
+
+    return data[['Player_ID', 'Knowledge_Text']]
+
+# Create a similarity-based retrieval function
+def query_knowledge_base(knowledge_base, query, vectorizer):
+    """Query the knowledge base using cosine similarity."""
+    query_vec = vectorizer.transform([query])
+    knowledge_vec = vectorizer.transform(knowledge_base['Knowledge_Text'])
+
+    # Compute cosine similarities
+    similarities = cosine_similarity(query_vec, knowledge_vec).flatten()
+
+    # Retrieve the most relevant rows
+    top_indices = np.argsort(similarities)[::-1][:5]  # Top 5 results
+    return knowledge_base.iloc[top_indices], similarities[top_indices]
+
+# Main pipeline with LLM integration and prompt engineering
+def main_pipeline(file_path, user_query):
+    """End-to-end pipeline for the RAG system with Llama3.2 and prompt engineering."""
+    # Preprocess data
+    data = preprocess_data(file_path)
+    knowledge_base = create_knowledge_base(data)
+
+    # Create TF-IDF Vectorizer
+    vectorizer = TfidfVectorizer()
+    vectorizer.fit(knowledge_base['Knowledge_Text'])
+
+    # Query the knowledge base
+    results, scores = query_knowledge_base(knowledge_base, user_query, vectorizer)
+
+    # Format retrieved knowledge for LLM input
+    retrieved_text = "\n".join(results['Knowledge_Text'].tolist())
+
+    # Use Llama3.2 for question answering with prompt engineering
+    llm = pipeline("text-generation", model="meta-llama/Llama-3.2-1B-Instruct")
+    prompt = (
+        f"You are an expert sports analyst. Based on the following training data, provide a detailed and insightful answer to the user's question. "
+        f"Always include relevant numerical data in your response. Limit your response to a maximum of 200 words.\n\n"
+        f"Training Data:\n{retrieved_text}\n\n"
+        f"User Question: {user_query}\n\nAnswer:"
+    )
+    response = llm(prompt, max_new_tokens=200, num_return_sequences=1)
+
+    # Extract the answer part only
+    answer = response[0]['generated_text'].split("Answer:", 1)[-1].strip()
+    return answer
+
+# Gradio interface
+def query_interface(file_path, user_query):
+    try:
+        result = main_pipeline(file_path.name, user_query)
+        return result
+    except Exception as e:
+        return str(e)
+
+# Launch Gradio app
+file_input = gr.File(label="Upload CSV File")
+text_input = gr.Textbox(label="Ask a Question", lines=2, placeholder="Enter your query here...")
+output = gr.Textbox(label="Answer")
+
+interface = gr.Interface(
+    fn=query_interface,
+    inputs=[file_input, text_input],
+    outputs=output,
+    title="RAG Training Data Query System",
+    description="Upload a CSV file containing training data and ask detailed questions about it."
+)
+
+interface.launch()