app.py

import pandas as pd
import gradio as gr
from catboost import CatBoostClassifier, Pool
from sklearn.model_selection import train_test_split, RandomizedSearchCV
from sklearn.metrics import accuracy_score
import joblib

# Sample Morningstar-like dataset
data = {
    'fund_rating': [5, 4, 3, 5, 2, 4, 1, 3, 5, 2],
    'fund_category': ['Equity', 'Debt', 'Balanced', 'Equity', 'Debt', 'Equity', 'Balanced', 'Debt', 'Equity', 'Debt'],
    'risk_appetite': ['High', 'Medium', 'Low', 'High', 'Low', 'Medium', 'Low', 'Medium', 'High', 'Low'],
    'expense_ratio': [0.5, 1.2, 1.0, 0.4, 1.5, 1.1, 1.3, 0.8, 0.3, 1.4],
    'user_age': [35, 45, 29, 50, 60, 40, 30, 28, 55, 62],
    'investment_goal_years': [5, 10, 3, 15, 2, 7, 4, 5, 20, 1],
    'will_invest': [1, 1, 0, 1, 0, 1, 0, 0, 1, 0]
}

# Convert to DataFrame
df = pd.DataFrame(data)
X = df.drop('will_invest', axis=1)
y = df['will_invest']
cat_features = ['fund_category', 'risk_appetite']

# Split data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
train_pool = Pool(X_train, y_train, cat_features=cat_features)

# Default model to start with
model = CatBoostClassifier(verbose=0)
model.fit(train_pool)

# Save initial model
joblib.dump(model, "catboost_model.pkl")

# Prediction function
def predict(fund_rating, fund_category, risk_appetite, expense_ratio, user_age, investment_goal_years):
    input_data = pd.DataFrame([[fund_rating, fund_category, risk_appetite, expense_ratio, user_age, investment_goal_years]],
                              columns=X.columns)
    loaded_model = joblib.load("catboost_model.pkl")
    return int(loaded_model.predict(input_data)[0])

# Tuning function
def tune_model():
    param_grid = {
        'depth': [4, 6, 8],
        'learning_rate': [0.01, 0.05, 0.1],
        'l2_leaf_reg': [1, 3, 5, 7],
        'n_estimators': [100, 200, 500],
        'bagging_temperature': [0, 0.5, 1],
    }
    search = RandomizedSearchCV(estimator=CatBoostClassifier(verbose=0),
                                param_distributions=param_grid,
                                scoring='accuracy',
                                cv=3,
                                n_iter=10,
                                random_state=42)
    search.fit(X_train, y_train, cat_features=cat_features)
    best_params = search.best_params_

    # Re-train model with best params
    tuned_model = CatBoostClassifier(**best_params, verbose=0)
    tuned_model.fit(train_pool)

    # Save tuned model
    joblib.dump(tuned_model, "catboost_model.pkl")

    y_pred = tuned_model.predict(X_test)
    acc = accuracy_score(y_test, y_pred)
    return f"Tuned! Accuracy: {acc:.2f}\nBest Params: {best_params}"

# Gradio Interface with Blocks
with gr.Blocks() as full_ui:
    gr.Markdown("## Mutual Fund Investment Classifier (CatBoost + Gradio)")

    with gr.Row():
        fund_rating = gr.Slider(1, 5, label="Fund Rating")
        fund_category = gr.Dropdown(choices=['Equity', 'Debt', 'Balanced'], label="Fund Category")
        risk_appetite = gr.Dropdown(choices=['High', 'Medium', 'Low'], label="Risk Appetite")

    with gr.Row():
        expense_ratio = gr.Slider(0.1, 2.0, step=0.1, label="Expense Ratio")
        user_age = gr.Slider(18, 80, label="User Age")
        investment_goal_years = gr.Slider(1, 30, label="Investment Goal (Years)")

    predict_btn = gr.Button("Predict")
    output = gr.Label()
    predict_btn.click(
        fn=predict,
        inputs=[fund_rating, fund_category, risk_appetite, expense_ratio, user_age, investment_goal_years],
        outputs=output
    )

    tune_btn = gr.Button("Tune Model")
    tune_out = gr.Textbox(label="Tuning Output")
    tune_btn.click(fn=tune_model, inputs=[], outputs=tune_out)

full_ui.launch()