Create rashomon_effect.py

my_pages/rashomon_effect.py

@@ -0,0 +1,92 @@
+# pages/rashomon_effect.py
+import streamlit as st
+import numpy as np
+import matplotlib.pyplot as plt
+
+st.title("Rashomon Effect")
+
+# Generate synthetic data
+np.random.seed(42)
+n_points = 100
+
+# Mostly correlated data
+annual_income = np.random.normal(50, 15, n_points)  # x-axis
+credit_score = annual_income + np.random.normal(0, 10, n_points)  # y-axis
+
+# Default labels: 1 = paid back (green), 0 = default (red)
+labels = (annual_income + credit_score > 120).astype(int)
+colors = np.where(labels == 1, 'green', 'red')
+
+# --- Function to draw scatter with optional decision boundary ---
+def draw_scatter(x, y, colors, boundary=None, boundary_type=None, extra_point=None):
+    fig, ax = plt.subplots()
+
+    # Background for decision boundary
+    if boundary is not None and boundary_type:
+        x_min, x_max = ax.get_xlim()
+        y_min, y_max = ax.get_ylim()
+
+        if boundary_type == "vertical":
+            ax.axvline(boundary, color="black", linestyle="--")
+            ax.fill_betweenx([y_min, y_max], x_min, boundary, color="red", alpha=0.1)
+            ax.fill_betweenx([y_min, y_max], boundary, x_max, color="green", alpha=0.1)
+
+        elif boundary_type == "horizontal":
+            ax.axhline(boundary, color="black", linestyle="--")
+            ax.fill_between([x_min, x_max], y_min, boundary, color="red", alpha=0.1)
+            ax.fill_between([x_min, x_max], boundary, y_max, color="green", alpha=0.1)
+
+    ax.scatter(x, y, c=colors, edgecolors='black')
+
+    if extra_point:
+        ax.scatter(*extra_point["coords"], c=extra_point["color"], s=150, edgecolors="black", marker="o")
+
+    ax.set_xlabel("Annual Income")
+    ax.set_ylabel("Credit Score")
+
+    return fig
+
+# --- Layout ---
+col_top = st.container()
+col_bottom = st.columns(2)
+
+# Draw top scatter (raw data)
+with col_top:
+    fig_top = draw_scatter(annual_income, credit_score, colors)
+    st.pyplot(fig_top)
+    st.markdown("⬇️ Data split into two different models ⬇️")
+
+# Session state to store choice
+if "chosen_boundary" not in st.session_state:
+    st.session_state.chosen_boundary = None
+
+# Bottom left: vertical decision boundary
+with col_bottom[0]:
+    fig_left = draw_scatter(annual_income, credit_score, colors, boundary=55, boundary_type="vertical")
+    st.pyplot(fig_left)
+    if st.button("Choose Vertical Model"):
+        st.session_state.chosen_boundary = "vertical"
+
+# Bottom right: horizontal decision boundary
+with col_bottom[1]:
+    fig_right = draw_scatter(annual_income, credit_score, colors, boundary=55, boundary_type="horizontal")
+    st.pyplot(fig_right)
+    if st.button("Choose Horizontal Model"):
+        st.session_state.chosen_boundary = "horizontal"
+
+# After choice: add extra individual
+if st.session_state.chosen_boundary:
+    st.markdown("---")
+    st.subheader("Your chosen model in action")
+
+    if st.session_state.chosen_boundary == "vertical":
+        extra = {"coords": (40, 80), "color": "blue"}
+        fig_choice = draw_scatter(annual_income, credit_score, colors, boundary=55, boundary_type="vertical", extra_point=extra)
+        st.pyplot(fig_choice)
+        st.warning("This person has a high credit score but low income. You rejected them. Why not choose a model that would approve them?")
+
+    elif st.session_state.chosen_boundary == "horizontal":
+        extra = {"coords": (80, 40), "color": "blue"}
+        fig_choice = draw_scatter(annual_income, credit_score, colors, boundary=55, boundary_type="horizontal", extra_point=extra)
+        st.pyplot(fig_choice)
+        st.warning("This person has high income but low credit score. You rejected them. Why not choose a model that would approve them?")