my_pages/rashomon_effect.py

import streamlit as st
import matplotlib.pyplot as plt
import numpy as np
from utils import add_navigation, add_instruction_text, add_red_text

plt.style.use('dark_background')

#### Setup data to plot
income = np.array([80, 85, 97, 91, 78, 102, 84, 88, 81, 40, 45, 51, 34, 47, 38, 39, 97, 91, 38, 32])
credit = np.array([970, 880, 1020, 910, 805, 800, 804, 708, 810, 370, 470, 309, 450, 304, 380, 501, 370, 301, 1080, 902])
labels = np.array([1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1])
colors = ['green' if label == 1 else 'red' for label in labels]

def plot_scatter(x, y, colors, title="", boundary_type=None, highlight_point=None):
    fig, ax = plt.subplots(figsize=(2, 2))
    ax.scatter(x, y, c=colors, alpha=0.6)
    ax.set_xlabel("Annual Income")
    ax.set_ylabel("Credit Score")
    # ax.set_title(title)

    fig.patch.set_alpha(0)
    ax.patch.set_alpha(0)

    # Decision boundary
    if boundary_type is not None:
        if boundary_type == "vertical":
            ax.axvline(65, color='blue')
            ax.fill_betweenx(np.arange(min(y), max(y)), 65, max(x), alpha=0.1, color='green')
            ax.fill_betweenx(np.arange(min(y), max(y)), min(x), 65, alpha=0.1, color='red')
        elif boundary_type == "horizontal":
            ax.axhline(650, color='blue')
            ax.fill_between(np.arange(min(x), max(x)), 650, max(y), alpha=0.1, color='green')
            ax.fill_between(np.arange(min(x), max(x)), min(y), 650, alpha=0.1, color='red')
        elif boundary_type == "slant":
            slope = -10.677966       # From (94, 350) and (35, 980)
            intercept = 1353.7288
            x_sorted = np.sort(x)
            y_line = slope * x_sorted + intercept
            ax.plot(x_sorted, y_line, color='blue')
            ax.fill_between(x_sorted, y_line, max(y), alpha=0.1, color='green')
            ax.fill_between(x_sorted, min(y), y_line, alpha=0.1, color='red')

    # Highlight specific point
    if highlight_point is not None:
        ax.scatter(*highlight_point, c='green', edgecolors='yellow', s=200, zorder=5, linewidths=4)

    ax.spines['right'].set_visible(False)
    ax.spines['top'].set_visible(False)
    ax.set_xticks([])
    ax.set_yticks([])
    
    return fig

def render():
    add_navigation("txt_rashomon_effect", "txt_rashomon_developer")

    add_instruction_text(
        """
        Consider the following data about individuals who did (green) or didn't (red) repay their loans. <br>
        Which model out of these three will you choose to judge loan applications?
        """
    )

    #### Rashomon Set Definition
    rashomon_set_message = """
        Multiple models achieving similar accuracy, i.e., multiple interpretations of the data, is known as the Rashomon effect.
        We call the models below part of a 'Rashomon set'.
    """
    add_red_text(rashomon_set_message)

    #### Plot three graphs to represent three models
    graph_selected, highlight_point = None, None
    if "graph_selected" in st.session_state:
        graph_selected = st.session_state.graph_selected
        highlight_point = st.session_state.highlight_point

    col1, col2, col3, col4, col5 = st.columns([0.5, 1, 1, 1, 0.5])
    with col2:
        st.pyplot(plot_scatter(income, credit, colors, boundary_type="vertical", highlight_point=highlight_point))
        st.markdown("Accuracy: 90%")
        if graph_selected=="vertical":
            button_click_v = st.button("Choose Model 1", type="primary")
        else:
            button_click_v = st.button("Choose Model 1")
        if button_click_v:
            st.session_state.highlight_point = (32, 902)
            st.session_state.graph_selected = "vertical"
            st.rerun()
    with col3:
        st.pyplot(plot_scatter(income, credit, colors, boundary_type="slant", highlight_point=highlight_point))
        st.markdown("Accuracy: 90%")
        if graph_selected=="slant":
            button_click_s = st.button("Choose Model 2", type="primary")
        else:
            button_click_s = st.button("Choose Model 2")
        if button_click_s:
            st.session_state.highlight_point = (32, 902)
            st.session_state.graph_selected = "slant"
            st.rerun()
    with col4:
        st.pyplot(plot_scatter(income, credit, colors, boundary_type="horizontal", highlight_point=highlight_point))
        st.markdown("Accuracy: 90%")
        if graph_selected=="horizontal":
            button_click_h = st.button("Choose Model 3", type="primary")
        else:
            button_click_h = st.button("Choose Model 3")
        if button_click_h:
            st.session_state.highlight_point = (97, 370)
            st.session_state.graph_selected = "horizontal"
            st.rerun()

    #### Multiplicity Definition    
    if "graph_selected" in st.session_state:
        multiplicity_message = """
            Because of your choice, the highlighted individual was rejected, but would have gotten loan under a different model.
            These conflicting predictions is multiplicity.<br><br>
            <b>Clearly, the choice of model directly impacts individuals!</b>
        """
        add_red_text(multiplicity_message)