Oracle/deepfundingoracle.py

"""
DeepFunding Oracle:
This script dynamically loads dependency data and for each repository URL:
  • Fetches GitHub features (stars, forks, watchers, open issues, pull requests, activity) using the GitHub API.
  • Uses the LLama model to analyze parent-child behavior (based on the fetched features and parent info)
    and returns a base weight (0-1) for the repository.
  • Trains a RandomForest regressor on these features (with the base weight as the target) to predict a final weight.
The output submission CSV has three columns: repo, parent, and final_weight.
"""
import base64
from io import StringIO
import os
import warnings
import csv
import re
import requests
import numpy as np
import pandas as pd
import time
import threading
import logging
import concurrent.futures
from concurrent.futures import ThreadPoolExecutor
import signal

from sklearn.pipeline import Pipeline
from tqdm import tqdm
import sys
import re
import json
import time

from sklearn.model_selection import train_test_split, GridSearchCV, RandomizedSearchCV
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error
from sklearn.preprocessing import StandardScaler
import matplotlib.pyplot as plt
import seaborn as sns
from scipy.special import log1p, expm1

from Oracle.SmolLM import SmolLM

warnings.filterwarnings("ignore")

# Configure logging to file and console
logging.basicConfig(
    handlers=[
        logging.FileHandler("deepfundingoracle.log"),
        logging.StreamHandler(sys.stdout)
    ],
    level=logging.INFO,
    format="%(asctime)s - %(levelname)s - %(message)s"
)

##############################
#  GitHub API helper: Fetch repository metrics
##############################
def fetch_repo_metrics(repo_url):
    """
    Fetch GitHub metrics (stars, forks, watchers, open issues, pull requests, and activity) given a repository URL.
    Assumes repo_url is in the form "https://github.com/owner/repo".
    Handles API failures and malformed URLs gracefully.
    """
    # Default values in case of failure
    default_metrics = {
        "stargazers_count": 0,
        "forks_count": 0, 
        "watchers_count": 0,
        "open_issues_count": 0,
        "pulls_count": 0,
        "activity": "",
        "contributors": 0,
        "dependencies_count": 0
    }
    
    try:
        # Extract owner and repo name
        m = re.search(r"github\.com/([^/]+)/([^/]+)", repo_url)
        if not m:
            print(f"[WARN] Malformed GitHub URL: {repo_url}")
            return default_metrics
            
        owner, repo_name = m.group(1), m.group(2)
        api_url = f"https://api.github.com/repos/{owner}/{repo_name}"
        headers = {}

        token = os.environ.get("GITHUB_API_TOKEN", "")
        if token:
            headers["Authorization"] = f"token {token}"
            
        # Fetch main repository data
        r = requests.get(api_url, headers=headers, timeout=10)
        if r.status_code == 200:
            data = r.json()
            metrics = {
                "stargazers_count": data.get("stargazers_count", 0),
                "forks_count": data.get("forks_count", 0),
                "watchers_count": data.get("watchers_count", 0),
                "open_issues_count": data.get("open_issues_count", 0),
                "activity": data.get("updated_at", ""),
                "owner": owner,
                "repo_name": repo_name,
                "dependencies_count": 0
            }
            
            # Try to fetch pull requests count
            try:
                pulls_url = f"{api_url}/pulls"
                pulls_resp = requests.get(pulls_url, headers=headers, timeout=5)
                metrics["pulls_count"] = len(pulls_resp.json()) if pulls_resp.status_code == 200 else 0
            except Exception as e:
                print(f"[WARN] Failed to fetch pulls for {repo_url}: {e}")
                metrics["pulls_count"] = 0
                
            # Try to fetch contributors count
            try:
                contributors_url = f"{api_url}/contributors"
                contributors_resp = requests.get(contributors_url, headers=headers, timeout=5)
                metrics["contributors"] = len(contributors_resp.json()) if contributors_resp.status_code == 200 else 0
            except Exception as e:
                print(f"[WARN] Failed to fetch contributors for {repo_url}: {e}")
                metrics["contributors"] = 0
                
            # Try to estimate dependencies from package files
            try:
                # Look for package.json for Node.js projects
                package_json_url = f"https://raw.githubusercontent.com/{owner}/{repo_name}/master/package.json"
                package_resp = requests.get(package_json_url, timeout=5)
                if package_resp.status_code == 200:
                    package_data = package_resp.json()
                    deps = package_data.get("dependencies", {})
                    dev_deps = package_data.get("devDependencies", {})
                    metrics["dependencies_count"] = len(deps) + len(dev_deps)
                else:
                    # Try requirements.txt for Python projects
                    req_txt_url = f"https://raw.githubusercontent.com/{owner}/{repo_name}/master/requirements.txt"
                    req_resp = requests.get(req_txt_url, timeout=5)
                    if req_resp.status_code == 200:
                        deps = [line for line in req_resp.text.split('\n') if line.strip() and not line.startswith('#')]
                        metrics["dependencies_count"] = len(deps)
            except Exception as e:
                print(f"[WARN] Failed to fetch dependencies for {repo_url}: {e}")
                metrics["dependencies_count"] = 0
                
            return metrics
        else:
            print(f"[ERROR] Failed to fetch data for {repo_url}: {r.status_code}")
            return default_metrics
    except Exception as e:
        print(f"[ERROR] Exception while fetching data for {repo_url}: {e}")
        return default_metrics

def fetch_github_features(df):
    """
    For each row, using the repo URL, call the GitHub API to fetch:
      stars, forks, watchers, open issues, pull requests, activity, and contributors count.
    Adds these as new columns to the DataFrame.
    """
    print("[INFO] Fetching GitHub features for repositories...")
    start_time = time.time()
    
    # Initialize lists for storing fetched data
    metrics_lists = {
        "stars": [],
        "forks": [],
        "watchers": [],
        "open_issues": [],
        "pulls": [],
        "activity": [],
        "contributors": [],
        "dependencies_count": []
    }

    cache = {}

    def get_metrics(repo_url):
        if repo_url in cache:
            print(f"[DEBUG] Cached GitHub data for {repo_url}: {cache[repo_url]}")
            return cache[repo_url]
        val = fetch_repo_metrics(repo_url)
        print(f"[DEBUG] Extracted GitHub data for {repo_url}: {val}")
        cache[repo_url] = val
        return val

    with concurrent.futures.ThreadPoolExecutor(max_workers=10) as executor:
        futures = {executor.submit(get_metrics, row['repo']): i for i, row in df.iterrows()}
        for fut in tqdm(concurrent.futures.as_completed(futures), total=len(futures), desc="Fetching metrics"):
            res = fut.result()
            metrics_lists["stars"].append(res.get("stargazers_count", 0))
            metrics_lists["forks"].append(res.get("forks_count", 0))
            metrics_lists["watchers"].append(res.get("watchers_count", 0))
            metrics_lists["open_issues"].append(res.get("open_issues_count", 0))
            metrics_lists["pulls"].append(res.get("pulls_count", 0))
            metrics_lists["activity"].append(res.get("activity", ""))
            metrics_lists["contributors"].append(res.get("contributors", 0))
            metrics_lists["dependencies_count"].append(res.get("dependencies_count", 0))

    # Add the fetched data to the DataFrame
    for key, values in metrics_lists.items():
        df[key] = values

    end_time = time.time()
    print(f"[INFO] GitHub features fetched successfully in {end_time - start_time:.2f} seconds.")
    return df

def calculate_fallback_weights(df):
    """
    Dynamically calculate fallback feature weights based on feature variance.
    """
    print("[INFO] Calculating fallback feature weights...")
    numeric_cols = ['stars', 'forks', 'watchers', 'open_issues', 'pulls', 'contributors', 'dependencies_count']
    # Filter to only include columns that exist in the DataFrame
    valid_cols = [col for col in numeric_cols if col in df.columns]
    
    # Create default weights
    default_weights = {
        'stars': 0.3,
        'forks': 0.2,
        'watchers': 0.2,
        'open_issues': 0.1,
        'pulls': 0.1,
        'contributors': 0.05,
        'dependencies_count': 0.05
    }
    
    # If any data exists, calculate variance-based weights
    if len(valid_cols) > 0 and df[valid_cols].sum().sum() > 0:
        # Calculate variance for each feature
        feature_variances = df[valid_cols].var()
        total_variance = feature_variances.sum()
        
        # If meaningful variance exists, use it for weights
        if total_variance > 0:
            weights = {col: var / total_variance for col, var in feature_variances.items()}
            # Normalize to ensure sum is 1.0
            sum_weights = sum(weights.values())
            if sum_weights > 0:
                weights = {k: v / sum_weights for k, v in weights.items()}
            return weights
    
    # Return default weights if we couldn't calculate meaningful ones
    print("[INFO] Using default fallback weights")
    return default_weights

##############################
#  Feature Extraction
##############################
def load_data(file):
    """
    Dynamically load the dependency data CSV from the uploaded file.
    Expects at least "repo" and "parent" columns.
    """
    try:
        print("[INFO] Loading data from uploaded file...")
        start_time = time.time()
        # Read the uploaded file directly into a DataFrame
        df = pd.read_csv(file)
        end_time = time.time()
        print(f"[INFO] Data loaded successfully in {end_time - start_time:.2f} seconds.")
        return df
    except Exception as e:
        print("[ERROR] Error loading data:", e)
        return None

def timeout_handler(signum, frame):
    raise TimeoutError("LLama model prediction timed out.")

def assign_base_weight(df, max_workers=32, llm_retries=2, llm_delay=0):
    """
    Assign base weights using a single LLM call to determine feature weights,
    and programmatically calculate repository weights.
    """
    print("[INFO] Starting optimized base weight assignment...", flush=True)
    logging.info("[INFO] Assigning base weights using optimized approach...")
    start_time = time.time()
    oracle = SmolLM()

    prompt = (
        "Can you Predict a weight in the range (0-1) for these GitHub features such as stars, forks, watchers, "
        "open_issues, pulls, activity, contributors based on their importance in determining the influence of a repository? "
        "Output the weights for each feature as text e.g.: "
        'stars: 0.3, forks: 0.2, watchers: 0.2, open_issues: 0.1, pulls: 0.1, activity: 0.05, contributors: 0.05'
    )
    feature_weights = None
    for attempt in range(llm_retries):
        try:
            response = oracle.predict(prompt, max_length=512, max_new_tokens=150)
            if not response or not response.strip():
                raise ValueError("Empty response from Oracle.")
            matches = re.findall(
                r'(stars|forks|watchers|open_issues|pulls|activity|contributors)\s*[:=]\s*([0-9]*\.?[0-9]+)',
                response, re.IGNORECASE)
            feature_weights = {k.lower(): float(v) for k, v in matches}
            if not feature_weights or len(feature_weights) < 7:
                raise ValueError("Could not extract all feature weights from response.")
            print(f"[INFO] Feature weights from LLM: {feature_weights}", flush=True)
            break
        except Exception as e:
            print(f"[ERROR] Oracle attempt {attempt+1} failed: {e}", flush=True)
            logging.error(f"[ERROR] Oracle attempt {attempt+1} failed: {e}")
            time.sleep(llm_delay)

    # Fallback mechanism: Calculate feature weights dynamically if LLM fails
    if feature_weights is None:
        print("[WARN] LLM failed to provide feature weights. Calculating fallback weights dynamically.")
        feature_weights = calculate_fallback_weights(df)
        print(f"[INFO] Fallback feature weights: {feature_weights}", flush=True)

    for feature in feature_weights.keys():
        if feature in df.columns:
            df[feature] = pd.to_numeric(df[feature], errors='coerce').fillna(0)

    def calculate_weight(row):
        weight = 0
        for feature, feature_weight in feature_weights.items():
            if feature in row:
                weight += row[feature] * feature_weight
        return weight

    df["base_weight_raw"] = df.apply(calculate_weight, axis=1)
    df["base_weight"] = df.groupby("parent")["base_weight_raw"].transform(
        lambda s: (s - s.min()) / (s.max() - s.min() if s.max() != s.min() else 1)
    )

    end_time = time.time()
    print(f"[INFO] Base weights assigned successfully in {end_time - start_time:.2f} seconds.", flush=True)
    logging.info(f"[INFO] Base weights assigned successfully in {end_time - start_time:.2f} seconds.")
    return df

def sanity_check_weights(df):
    """
    Sanity-checks LLM weights by comparing them with other metrics.
    """
    print("[INFO] Performing sanity check on LLM weights...")
    df["sanity_check_weight"] = (df["stars"] + df["forks"] + df["watchers"]) / 3
    df["ensemble_weight"] = (df["base_weight"] + df["sanity_check_weight"]) / 2
    print("[INFO] Sanity check and ensemble weights added.")
    return df

def visualize_feature_distributions(df):
    """
    Visualizes feature distributions and correlations.
    """
    print("[INFO] Visualizing feature distributions and correlations...")
    if not plt or not sns:
        print("[ERROR] Matplotlib or Seaborn not available for visualization.")
        return
    numeric_cols = df.select_dtypes(include=[np.number]).columns

    # Plot feature distributions
    df[numeric_cols].hist(bins=20, figsize=(15, 10), color="skyblue", edgecolor="black")
    plt.suptitle("Feature Distributions", fontsize=16)
    plt.show()

    # Plot feature correlations
    correlation_matrix = df[numeric_cols].corr()
    plt.figure(figsize=(12, 8))
    sns.heatmap(correlation_matrix, annot=True, cmap="coolwarm", fmt=".2f", linewidths=0.5)
    plt.title("Feature Correlation Matrix", fontsize=16)
    plt.show()


def normalize_and_clip_weights(df, group_col="parent", weight_col="final_weight"):
    """
    Ensures weights are non-negative and sum to 1 per group.
    """
    if df is None:
        raise ValueError("DataFrame is None, cannot normalize weights.")
    if weight_col not in df.columns:
        raise KeyError(f"`{weight_col}` column not found in DataFrame.")
    # Clip negatives
    df[weight_col] = df[weight_col].clip(lower=0)
    # Normalize within each group
    def normalize_group(x):
        total = x.sum()
        if total > 0:
            return x / total
        return np.ones_like(x) / len(x)
    df[weight_col] = df.groupby(group_col)[weight_col].transform(normalize_group)
    return df

def normalize_funding(df):
    """
    Normalize funding weights for child repositories grouped by parent.
    """
    print("[INFO] Normalizing funding weights...", flush=True)
    if df is None or df.empty:
        print("[WARN] Skipping normalization: DataFrame is None or empty.", flush=True)
        return df
    df = normalize_and_clip_weights(df)
    print("[INFO] Funding weights normalized successfully.", flush=True)
    return df

def prepare_dataset(file):
    print("[INFO] Starting dataset preparation...")
    start_time = time.time()
    df = load_data(file)
    if df is None:
        raise ValueError("Failed to load data.")
    if not {"repo", "parent"}.issubset(df.columns):
        raise ValueError("Input CSV must contain 'repo' and 'parent' columns.")
    print("[INFO] Fetching GitHub features...")
    df = fetch_github_features(df)
    print("[INFO] GitHub features fetched successfully.")
    print("[INFO] Cleaning data...")
    df = clean_data(df)
    print("[INFO] Data cleaned successfully.")
    print("[INFO] Assigning base weights using LLama model...")
    df = assign_base_weight(df)
    df = sanity_check_weights(df)  # Add sanity-check and ensemble weights
    df = train_predict_weight(df)
    if df is not None and not df.empty:
        visualize_feature_distributions(df)
    else:
        print("[WARN] DataFrame is empty after processing. Skipping visualization.")
    df = normalize_funding(df)
    end_time = time.time()
    print(f"[INFO] Dataset preparation completed in {end_time - start_time:.2f} seconds.")
    return df


##############################
#  Data Cleaning
##############################
def clean_data(df):
    """
    Cleans the input DataFrame by handling missing values and removing outliers.
    """
    # Impute missing values
    df.fillna(df.median(numeric_only=True), inplace=True)

    # Remove extreme outliers using quantiles
    for col in df.select_dtypes(include=[np.number]).columns:
        q1 = df[col].quantile(0.25)
        q3 = df[col].quantile(0.75)
        iqr = q3 - q1
        lower_bound = q1 - 1.5 * iqr
        upper_bound = q3 + 1.5 * iqr
        df[col] = df[col].clip(lower=lower_bound, upper=upper_bound)

    return df


##############################
#  Feature Validation and Scaling
##############################
def validate_features(df):
    """
    Validates and scales features to ensure they are meaningful for model training.
    """
    print("[INFO] Validating and scaling features...")
    numeric_cols = df.select_dtypes(include=[np.number]).columns
    for col in numeric_cols:
        df[col]= log1p(df[col].clip(lower=0))
    scaler = StandardScaler()
    # Scale numeric features
    df[numeric_cols] = scaler.fit_transform(df[numeric_cols])
    print("[INFO] Features scaled successfully.")
    return df

def validate_target(df):
    """
    Validates the target variable to ensure it has sufficient variance.
    If variance is insufficient, adds small random noise to create variance.
    """
    print("[INFO] Validating target variable 'base_weight'...")
    target = "base_weight"
    if target not in df.columns:
        raise ValueError(f"Target variable '{target}' not found in DataFrame.")

    variance = df[target].var()
    print(f"[DEBUG] Target variable variance: {variance}")
    if variance < 1e-6:
        print("[WARN] Target variable has insufficient variance. Adding small random noise...")
        # Add small random noise to introduce variance
        np.random.seed(42)  # For reproducibility
        noise = np.random.normal(0.5, 0.1, size=len(df))
        df[target] = noise
        print(f"[INFO] New target variable variance: {df[target].var()}")
    return df


##############################
#  RandomForest Regression
##############################
def train_predict_weight(df):
    """
    Trains a RandomForestRegressor with hyperparameter tuning and evaluates the model.
    """
    print("[INFO] Starting weight prediction with hyperparameter tuning...", flush=True)
    start_time = time.time()
    target = "base_weight"
    feature_cols = [col for col in df.select_dtypes(include=[np.number]).columns if col not in ["base_weight", "final_weight","base_weight_raw"]]

    X = df[feature_cols].fillna(0)
    y = df[target]


    # Remove rows with NaN values
    mask = X.notna().all(axis=1) & y.notna()
    X,y = X[mask], y[mask]

    # Check for sufficient data and variance
    if X.shape[0] < 5 or y.nunique() <=1:
        print("[WARN] Not enough data or variance for model training. Using base weights directly.")
        df["final_weight"] = df[target]
        return normalize_and_clip_weights(df)

    # log1p transform target
    y_log = log1p(y)

    # Split data into train/test sets
    X_train, X_test, y_train_log, y_test_log = train_test_split(X, y_log, test_size=0.2, random_state=42)

    pipeline = Pipeline([
        ("rf", RandomForestRegressor(random_state=42))
    ])
    # Hyperparameter tuning using GridSearchCV
    param_dist = {
        "rf__n_estimators": [100, 300, 500, 800, 1000],
        "rf__max_depth": [None, 20, 30, 40],
        "rf__min_samples_split": [2, 5, 10],
        "rf__min_samples_leaf": [1, 2, 4],
        "rf__max_features": ["auto", "sqrt"],
    }
    search = RandomizedSearchCV(
        pipeline,
        param_distributions=param_dist,
        n_iter=50,
        cv=10,
        scoring="neg_root_mean_squared_error",
        verbose=2,
        n_jobs=-1,
        random_state=42
    )
    search.fit(X_train, y_train_log)
    best_model = search.best_estimator_

    #Predict on test, invert transform
    y_pred_test_log = best_model.predict(X_test)
    y_pred_test = expm1(y_pred_test_log)
    y_true_test = expm1(y_test_log)
    mse = mean_squared_error(y_true_test, y_pred_test)
    print(f"[INFO] Test MSE after RandomizedSearch: {mse:.4f}", flush=True)
    # Predict on full dataset and invert
    df["final_weight"] = expm1(best_model.predict(df[feature_cols]))
    df = normalize_and_clip_weights(df)
    end_time = time.time()
    print(f"[INFO] Weight prediction completed in {end_time - start_time:.2f} seconds.", flush=True)
    return df


##############################
# CSV Output
##############################
def create_submission_csv(df, output_filename="submission.csv"):
    print(f"[INFO] Writing results to {output_filename}...", flush=True)
    required_cols = ["repo", "parent", "final_weight"]
    submission_df = df[required_cols]
    submission_df.to_csv(output_filename, index=False)
    print(f"[INFO] Results written to {output_filename}.", flush=True)
    return output_filename

# Removed Gradio UI code from this file to ensure modular workflow.
# This file now focuses solely on data processing and prediction.

if __name__ == "__main__":
    input_file = "input.csv"  # Replace with the actual input file path
    output_file = "submission.csv"

    print("[INFO] Preparing dataset...")
    df = prepare_dataset(input_file)

    print("[INFO] Creating submission CSV...")
    create_submission_csv(df, output_file)
    print("[INFO] Process completed successfully.")