app.py

#!/usr/bin/env python
# coding: utf-8

import joblib
import pandas as pd
import gradio as gr
import matplotlib.pyplot as plt

plt.rcParams["figure.dpi"] = 100

# ============================================================
# Fonts
# ============================================================
plt.rcParams["font.family"] = "Arial"
plt.rcParams["font.weight"] = "black"
plt.rcParams["font.size"] = 10

TITLE_FONTSIZE = 14
TEXT_FONTSIZE = 10

# ============================================================
# Colors
# ============================================================
RED = "#d62728"
BLUE = "#1f77b4"

# ============================================================
# Class meaning
# ============================================================
LEAVE_CLASS = 1
STAY_CLASS = 0

# =========================
# Load model
# =========================
model = joblib.load("final.joblib")

FEATURES = [
    "Engagement",
    "SupportiveGM",
    "ManagementLevel",
    "WellBeing",
    "Voice",
    "DecisionAutonomy",
    "Workload",
    "WorkEnvironment",
]

# =========================
# Cluster anchors
# =========================
CLUSTER_1 = {
    "Voice": 4.84,
    "DecisionAutonomy": 4.90,
    "Workload": 4.72,
    "WellBeing": 4.8397,
    "WorkEnvironment": 4.8858,
    "SupportiveGM": 4.8583,
    "Engagement": 4.9324,
}

CLUSTER_2 = {
    "Voice": 3.94,
    "DecisionAutonomy": 4.24,
    "Workload": 3.76,
    "WellBeing": 4.0251,
    "WorkEnvironment": 4.1484,
    "SupportiveGM": 4.1275,
    "Engagement": 4.2828,
}

CLUSTER_3 = {
    "Voice": 2.39,
    "DecisionAutonomy": 3.55,
    "Workload": 2.68,
    "WellBeing": 3.0299,
    "WorkEnvironment": 3.4537,
    "SupportiveGM": 3.2208,
    "Engagement": 3.3909,
}

ALL_DRIVER_VARS = [
    "Engagement",
    "SupportiveGM",
    "WellBeing",
    "WorkEnvironment",
    "Voice",
    "DecisionAutonomy",
    "Workload",
]

ALL_DRIVER_LABELS = [
    "Engagement",
    "Supportive GM",
    "Well-Being",
    "Work Environment",
    "Voice",
    "Decision Autonomy",
    "Workload",
]

GOAL_AVG = {v: CLUSTER_1[v] for v in ALL_DRIVER_VARS}


# ============================================================
# Helpers
# ============================================================
def clamp_1_5(x):
    return max(1.0, min(5.0, float(x)))


def build_X(vals):
    row = {f: vals[f] for f in FEATURES}
    return pd.DataFrame([[row[f] for f in FEATURES]], columns=FEATURES)


def prob_leave_and_stay(X):
    probs = model.predict_proba(X)[0]
    classes = list(model.classes_)
    p_leave = float(probs[classes.index(LEAVE_CLASS)])
    p_stay = float(probs[classes.index(STAY_CLASS)])
    return p_leave, p_stay


def hex_to_rgb01(h):
    h = h.lstrip("#")
    return (
        int(h[0:2], 16) / 255.0,
        int(h[2:4], 16) / 255.0,
        int(h[4:6], 16) / 255.0,
    )


# ============================================================
# Donut chart
# ============================================================
def make_turnover_donut(p_leave, p_stay):

    fig, ax = plt.subplots(figsize=(8, 3))

    ax.pie(
        [p_leave, p_stay],
        startangle=90,
        colors=[RED, BLUE],
        wedgeprops=dict(width=0.35, edgecolor="white"),
    )

    ax.text(
        0,
        0,
        f"{p_leave*100:.0f}%",
        ha="center",
        va="center",
        fontsize=18,
        color=RED,
        fontweight="black",
    )

    ax.set_title("Turnover Risk", fontsize=TITLE_FONTSIZE)

    plt.close(fig)
    return fig


# ============================================================
# Driver chart
# ============================================================
def make_driver_plot(vals):

    values = [vals[v] for v in ALL_DRIVER_VARS]
    goals = [GOAL_AVG[v] for v in ALL_DRIVER_VARS]

    fig, ax = plt.subplots(figsize=(12,3))

    bars = ax.bar(range(len(values)), values)

    for i,b in enumerate(bars):
        y = goals[i]
        ax.plot([b.get_x(), b.get_x()+b.get_width()], [y,y], linestyle="--")

    ax.set_xticks(range(len(values)))
    ax.set_xticklabels(ALL_DRIVER_LABELS)

    ax.set_ylim(1,5.4)
    ax.set_title("Average of key drivers")

    plt.close(fig)
    return fig


# ============================================================
# SHAP (original waterfall restored, red forced)
# ============================================================
def make_catboost_shap_plot(X):

    import shap
    from catboost import Pool

    try:

        shap.plots.colors.red_rgb = hex_to_rgb01(RED)
        shap.plots.colors.blue_rgb = hex_to_rgb01(BLUE)

        pool = Pool(X)
        shap_vals = model.get_feature_importance(pool, type="ShapValues")

        classes = list(model.classes_)
        class_idx = classes.index(LEAVE_CLASS)

        if shap_vals.ndim == 3:
            base = float(shap_vals[0, -1, class_idx])
            values = shap_vals[0, :-1, class_idx]
        else:
            base = float(shap_vals[0, -1])
            values = shap_vals[0, :-1]

        feature_names = list(X.columns)
        data_row = X.iloc[0].values

        exp = shap.Explanation(
            values=values,
            base_values=base,
            data=data_row,
            feature_names=feature_names,
        )

        shap.plots.waterfall(exp, max_display=8, show=False)

        fig = plt.gcf()
        fig.set_size_inches(8,3)

        plt.tight_layout()
        plt.close(fig)

        return fig

    except Exception as e:

        fig, ax = plt.subplots()
        ax.text(0.5,0.5,"SHAP unavailable",ha="center")
        plt.close(fig)
        return fig


# ============================================================
# Predict
# ============================================================
def predict(Engagement,SupportiveGM,WellBeing,WorkEnvironment,Voice,DecisionAutonomy,Workload):

    vals = {
        "Engagement": clamp_1_5(Engagement),
        "SupportiveGM": clamp_1_5(SupportiveGM),
        "WellBeing": clamp_1_5(WellBeing),
        "WorkEnvironment": clamp_1_5(WorkEnvironment),
        "Voice": clamp_1_5(Voice),
        "DecisionAutonomy": clamp_1_5(DecisionAutonomy),
        "Workload": clamp_1_5(Workload),
        "ManagementLevel": 2,
    }

    X = build_X(vals)

    p_leave, p_stay = prob_leave_and_stay(X)

    donut = make_turnover_donut(p_leave,p_stay)
    shap = make_catboost_shap_plot(X)
    drivers = make_driver_plot(vals)

    return donut, shap, drivers


# ============================================================
# Buttons
# ============================================================
def load_risk():

    target = {v:(CLUSTER_2[v]+CLUSTER_3[v])/2 for v in ALL_DRIVER_VARS}

    return (
        target["Engagement"],
        target["SupportiveGM"],
        target["WellBeing"],
        target["WorkEnvironment"],
        target["Voice"],
        target["DecisionAutonomy"],
        target["Workload"],
        *predict(**target)
    )


def apply_recommendation():

    target = {v:CLUSTER_1[v] for v in ALL_DRIVER_VARS}

    return (
        target["Engagement"],
        target["SupportiveGM"],
        target["WellBeing"],
        target["WorkEnvironment"],
        target["Voice"],
        target["DecisionAutonomy"],
        target["Workload"],
        *predict(**target)
    )


def hilton_heroes():
    return apply_recommendation()


# ============================================================
# CSS
# ============================================================
CSS = f"""
#btn_risk button {{ background:{RED}; color:white; }}
#btn_heroes button {{ background:#0a5eb8; color:white; }}
#btn_reco button {{ background:{BLUE}; color:white; }}
"""


# ============================================================
# UI
# ============================================================
with gr.Blocks(css=CSS) as demo:

    gr.Markdown("## Predicting Intent to Stay")

    with gr.Row():
        btn_risk = gr.Button("Immediate and Silent Risk", elem_id="btn_risk")
        btn_heroes = gr.Button("Hilton Heroes", elem_id="btn_heroes")
        btn_reco = gr.Button("Apply Recommendation", elem_id="btn_reco")

    with gr.Row():

        with gr.Column():

            Engagement = gr.Slider(1,5,value=3,label="Engagement")
            SupportiveGM = gr.Slider(1,5,value=3,label="Supportive GM")
            WellBeing = gr.Slider(1,5,value=3,label="Well Being")
            WorkEnvironment = gr.Slider(1,5,value=3,label="Work Environment")
            Voice = gr.Slider(1,5,value=3,label="Voice")
            DecisionAutonomy = gr.Slider(1,5,value=3,label="Decision Autonomy")
            Workload = gr.Slider(1,5,value=3,label="Workload")

            btn_predict = gr.Button("Predict")

        with gr.Column():

            donut_plot = gr.Plot()
            shap_plot = gr.Plot()

    drivers_plot = gr.Plot()

    sliders = [
        Engagement,
        SupportiveGM,
        WellBeing,
        WorkEnvironment,
        Voice,
        DecisionAutonomy,
        Workload,
    ]

    btn_predict.click(
        predict,
        sliders,
        [donut_plot,shap_plot,drivers_plot]
    )

    btn_risk.click(
        load_risk,
        [],
        sliders+[donut_plot,shap_plot,drivers_plot]
    )

    btn_heroes.click(
        hilton_heroes,
        [],
        sliders+[donut_plot,shap_plot,drivers_plot]
    )

    btn_reco.click(
        apply_recommendation,
        [],
        sliders+[donut_plot,shap_plot,drivers_plot]
    )
 
demo.launch()