utils/plot_gains.py

import pandas as pd
import streamlit as st
import plotly.graph_objects as go #type: ignore

def plot_gains(y_true, y_probs):
    # Build and sort dataframe
    df = pd.DataFrame({
        'Actual': y_true,
        'Predicted': y_probs
    }).sort_values(by='Predicted', ascending=False).reset_index(drop=True)

    # Compute cumulative gain
    df['Cumulative Percent'] = df['Actual'].cumsum() / df['Actual'].sum()
    df['Percent of Data'] = (df.index + 1) / len(df)

    # Compute K-stat (max distance from curve)
    df['ks_stat'] = df['Cumulative Percent'] - df['Percent of Data']
    ks_value = df['ks_stat'].max()
    ks_idx = df['ks_stat'].idxmax()
    cum_percent = df['Cumulative Percent'][ks_idx]
    data_percent = df['Percent of Data'][ks_idx]

    # Plotly figure
    fig = go.Figure()

    # Model Gains Curve
    fig.add_trace(go.Scatter(
        x=df['Percent of Data'],
        y=df['Cumulative Percent'],
        mode='lines',
        name='Model Gains Curve',
        line=dict(width=3)
    ))

    # Random baseline
    fig.add_trace(go.Scatter(
        x=[0, 1],
        y=[0, 1],
        mode='lines',
        name='Random Baseline',
        line=dict(width=2, dash='dash', color='gray')
    ))

    fig.add_annotation(
        x=data_percent,
        y=cum_percent,
        text=f'Best Returns: {data_percent*100:.2f}%'
    )

    fig.update_layout(
        title="Gains Curve",
        xaxis_title="Percent of Data",
        yaxis_title="Percent of Total Positive Cases Captured",
        template="plotly_white",
        height=450,
        legend=dict(yanchor="bottom", y=0, xanchor="right", x=1)
    )

    st.session_state.ks_value = ks_value
    st.session_state.peak_gains = round(data_percent * 100, 2)
    st.session_state.percent_data = round(cum_percent * 100, 2)

    return fig, data_percent, cum_percent