app.py

import pickle
import pandas as pd
import shap
from shap.plots._force_matplotlib import draw_additive_plot
import gradio as gr
import numpy as np
import matplotlib.pyplot as plt

# load the model from disk
loaded_model = pickle.load(open("h22_xgb.pkl", 'rb'))

# Setup SHAP
explainer = shap.Explainer(loaded_model) # PLEASE DO NOT CHANGE THIS.

# Create the main function for server
def main_func(ValueDiversity,AdequateResources,Voice,GrowthAdvancement,Workload,WorkLifeBalance):
    new_row = pd.DataFrame.from_dict({'ValueDiversity':ValueDiversity,'AdequateResources':AdequateResources,
              'Voice':Voice,'GrowthAdvancement':GrowthAdvancement,'Workload':Workload,
              'WorkLifeBalance':WorkLifeBalance}, orient = 'index').transpose()
    
    prob = loaded_model.predict_proba(new_row)
    
    shap_values = explainer(new_row)
    # plot = shap.force_plot(shap_values[0], matplotlib=True, figsize=(30,30), show=False)
    # plot = shap.plots.waterfall(shap_values[0], max_display=6, show=False)
    plot = shap.plots.bar(shap_values[0], max_display=6, order=shap.Explanation.abs, show_data='auto', show=False)

    plt.tight_layout()
    local_plot = plt.gcf()
    plt.close()
    
    return {"Leave": float(prob[0][0]), "Stay": 1-float(prob[0][0])}, local_plot

# Create the UI
title = "**Employee Turnover Predictor & Interpreter** 🪐"
description1 = """
This app takes six inputs about employees' satisfaction with different aspects of their work (such as work-life balance, ...) and predicts whether the employee intends to stay with the employer or leave. There are two outputs from the app: 1- the predicted probability of stay or leave, 2- Shapley's force-plot which visualizes the extent to which each factor impacts the stay/ leave prediction.✨   
"""

description2 = """
To use the app, click on one of the examples, or adjust the values of the six employee satisfaction factors, and click on Analyze. 🤞
""" 

with gr.Blocks(title=title) as demo:
    gr.Markdown(f"## {title}")
#    gr.Markdown("""![marketing](file/marketing.jpg)""")
    gr.Markdown(description1)
    gr.Markdown("""---""")
    gr.Markdown(description2)
    gr.Markdown("""---""")
    ValueDiversity = gr.Slider(label="ValueDiversity Score", minimum=1, maximum=5, value=4, step=1)
    AdequateResources = gr.Slider(label="AdequateResources Score", minimum=1, maximum=5, value=4, step=1)
    Voice = gr.Slider(label="Voice Score", minimum=1, maximum=5, value=4, step=1)
    GrowthAdvancement = gr.Slider(label="GrowthAdvancement Score", minimum=1, maximum=5, value=4, step=1)
    Workload = gr.Slider(label="Workload Score", minimum=1, maximum=5, value=4, step=1)
    WorkLifeBalance = gr.Slider(label="WorkLifeBalance Score", minimum=1, maximum=5, value=4, step=1)
    submit_btn = gr.Button("Analyze")

    with gr.Column(visible=True) as output_col:
        label = gr.Label(label = "Predicted Label")
        local_plot = gr.Plot(label = 'Shap:')

    submit_btn.click(
        main_func,
        [ValueDiversity,AdequateResources,Voice,GrowthAdvancement,Workload,WorkLifeBalance],
        [label,local_plot], api_name="Employee_Turnover"
    )
    
    gr.Markdown("### Click on any of the examples below to see how it works:")
    gr.Examples([[4,4,4,4,5,5], [5,4,5,4,4,4]], [ValueDiversity,AdequateResources,Voice,GrowthAdvancement,Workload,WorkLifeBalance], [label,local_plot], main_func, cache_examples=True)

demo.launch()