app.py

import pickle
import pandas as pd
import shap
import gradio as gr
import numpy as np
import matplotlib.pyplot as plt

# Load the XGBoost model from Pickle
loaded_model = pickle.load(open("h22_xgb_Final(2).pkl", 'rb'))

# Setup SHAP Explainer for XGBoost (Do not change this)
explainer = shap.Explainer(loaded_model)

def safe_convert(value, default, min_val, max_val):
    try:
        num = float(value)
        return max(min_val, min(num, max_val))  # Ensure within range
    except (TypeError, ValueError):
        return default  # Use default if conversion fails

# Create the main function for the model
def main_func(Department, ChainScale, SupportiveGM, Merit, LearningDevelopment, WorkEnvironment, Engagement, WellBeing):
    
    
    ChainScale_mapping = {
        'Luxury': 1,
        'Upper Midscale': 2,
        'Upper Upscale': 3,
        'Upscale': 4,
        'Independent': 5,
    }
    
    department_mapping = {
        "Guest Services": 1,
        "Food and Beverage": 2,
        "Housekeeping": 3,
        "Front Office Operations": 4,
        "Guest Activities": 5,
    }

    
    LearningDevelopment = safe_convert(LearningDevelopment, 3.0, 1, 5)
    SupportiveGM = safe_convert(SupportiveGM, 3.0, 1, 5)
    Merit = safe_convert(Merit, 3.0, 1, 5)
    WorkEnvironment = safe_convert(WorkEnvironment, 3.0, 1, 5)
    Engagement = safe_convert(Engagement, 3.0, 1, 5)
    WellBeing = safe_convert(WellBeing, 3.0, 1, 5)

    
    new_row = pd.DataFrame({
        'SupportiveGM': [SupportiveGM],
        'Merit': [Merit],
        'LearningDevelopment': [LearningDevelopment],
        'WorkEnvironment': [WorkEnvironment],
        'Engagement': [Engagement],
        'WellBeing': [WellBeing]
    }).astype(float)

    # Predict probabilities
    prob = loaded_model.predict_proba(new_row) 
    
    if prob.shape[1] == 2:
        leave_prob = float(prob[0][0])
        stay_prob = float(prob[0][1])
    else:
        leave_prob = float(prob[0])
        stay_prob = 1 - leave_prob

    # Generate SHAP values
    shap_values = explainer(new_row)

    fig, ax = plt.subplots(figsize=(8, 4))
    shap.waterfall_plot(shap.Explanation(
        values=shap_values.values[0],
        base_values=shap_values.base_values[0],
        data=new_row.iloc[0]
    ))
    plt.tight_layout()
    local_plot = plt.gcf()
    plt.close()

    return {"Leave": leave_prob, "Stay": stay_prob}, local_plot

# Create the UI
title = "**Mod 3 Team 5: Employee Turnover Predictor & Interpreter**"
description1 = """
This app evaluates six key factors affecting employee satisfaction—Supportive GM, Merit, Learning & Development, Work Environment, Engagement, and Well-Being—to predict whether an employee is likely to stay with Hilton or leave.
The app provides two key outputs:

**Predicted Probability**  
A likelihood score indicating whether an employee will stay or leave.

**SHAP Force Plot**  
A dynamic visualization that illustrates how each factor influences the prediction, helping to pinpoint the most impactful drivers of employee retention.
Designed for HR teams at both departmental and hotel chain levels, this tool delivers data-driven insights to improve employee experience and retention strategies across Hilton properties.
"""

description2 = """
To use the app, adjust the values of the employee satisfaction factors and click on Analyze.
"""

with gr.Blocks(title=title) as demo:
    gr.Markdown(f"## {title}")
    gr.Markdown(description1)
    gr.Markdown("""---""")
    gr.Markdown(description2)
    gr.Markdown("""---""")
    
    with gr.Row():      
        with gr.Column():
            Department = gr.Radio(
                ["Guest Services", "Food and Beverage", "Housekeeping", "Front Office Operations", "Guest Activities"],
                label="Department", 
                value="Guest Services"
            )
            ChainScale = gr.Dropdown(
                ["Luxury", "Upper Midscale", "Upper Upscale", "Upscale", "Independent"],
                label="ChainScale (Example)", 
                value="Upper Upscale"
            )
            SupportiveGM = gr.Slider(
                label="SupportiveGM Score", minimum=1, maximum=5, value=4, step=0.1,
                interactive=True
            )
            Merit = gr.Slider(
                label="Merit Score", minimum=1, maximum=5, value=4, step=0.1,
                interactive=True
            )
            LearningDevelopment = gr.Slider(
                label="Learning and Development Score", minimum=1, maximum=5, value=4, step=0.1,
                interactive=True
            )
            WorkEnvironment = gr.Slider(
                label="Work Environment Score", minimum=1, maximum=5, value=4, step=0.1,
                interactive=True
            )
            Engagement = gr.Slider(
                label="Engagement Score", minimum=1, maximum=5, value=4, step=0.1,
                interactive=True
            )
            WellBeing = gr.Slider(
                label="Well-Being Score", minimum=1, maximum=5, value=4, step=0.1,
                interactive=True
            )
            submit_btn = gr.Button("Analyze")
        
        with gr.Column(visible=True, scale=1, min_width=600) as output_col:
            label = gr.Label(label="Predicted Intent to Stay vs Leave")
            local_plot = gr.Plot(label='SHAP Waterfall Analysis')

    submit_btn.click(
        main_func,
        [Department, ChainScale, SupportiveGM, Merit, LearningDevelopment, WorkEnvironment, Engagement, WellBeing],
        [label, local_plot],
        api_name="Employee_Turnover"
    )
    
    gr.Markdown("### Click on any of the examples below to see how it works:")
    
    gr.Examples(
        [
            ["Guest Services", "Upper Upscale", 4.1, 3.7, 3.9, 4.2, 4.4, 4.3],  
            ["Food and Beverage", "Upper Upscale", 3.9, 3.7, 4.1, 4.3, 4.5, 4.4],  
            ["Housekeeping", "Upper Upscale", 4.3, 4.0, 4.3, 4.4, 4.5, 4.4],  
        ],
        [Department, ChainScale, SupportiveGM, Merit, LearningDevelopment, WorkEnvironment, Engagement, WellBeing],
        [label, local_plot],
        main_func,
        cache_examples=True
    )
demo.launch()