Create rashomon_developer.py

my_pages/rashomon_developer.py

@@ -0,0 +1,81 @@
+import streamlit as st
+import matplotlib.pyplot as plt
+import numpy as np
+from utils import add_navigation, add_instruction_text
+from rashomon_effect import plot_scatter
+from rashomon_effect import income, credit, labels, colors
+
+plt.style.use('dark_background')
+
+def render():
+    add_navigation("txt_rashomon_effect", "txt_developer_decisions")
+
+    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'.
+    """
+    st.markdown(
+        f"<div style='text-align:center; color:#c0392b; font-size:20px; margin:0 0;'>{rashomon_set_message}</div>",
+        unsafe_allow_html=True,
+    )
+
+    #### 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>
+        """
+        st.markdown(
+            f"<div style='text-align:center; color:#c0392b; font-size:20px; margin:0;'>{multiplicity_message}</div>",
+            unsafe_allow_html=True,
+        )