Update my_pages/ica.py

my_pages/ica.py

@@ -11,10 +11,11 @@ def render():
     add_instruction_text(
         """
         Explore the intention-convention-arbitrariness (ICA) framework.<br>
-        Use the sliders to adjust the three dimensions simultaneously.
+        Use the sliders to adjust the three dimensions.
         """
     )
 
+    # Initialize weights
     if "weights" not in st.session_state:
         st.session_state.weights = {
             "Intentional": 0.33,
@@ -22,27 +23,56 @@ def render():
             "Arbitrary": 0.34
         }
 
+    # Keep track of previous weights
+    if "prev_weights" not in st.session_state:
+        st.session_state.prev_weights = st.session_state.weights.copy()
+
     w = st.session_state.weights
+    prev_w = st.session_state.prev_weights
 
     # --- Three sliders ---
     col1, col2, col3 = st.columns(3)
     with col1:
-        i = st.slider("Intentional", 0.0, 1.0, w["Intentional"], 0.01)
+        i_new = st.slider("Intentional", 0.0, 1.0, w["Intentional"], 0.01)
     with col2:
-        c = st.slider("Conventional", 0.0, 1.0, w["Conventional"], 0.01)
+        c_new = st.slider("Conventional", 0.0, 1.0, w["Conventional"], 0.01)
     with col3:
-        a = st.slider("Arbitrary", 0.0, 1.0, w["Arbitrary"], 0.01)
-
-    # Normalize to sum = 1
-    total = i + c + a
-    if total > 0:
-        w["Intentional"] = round(i / total, 4)
-        w["Conventional"] = round(c / total, 4)
-        w["Arbitrary"] = round(a / total, 4)
-    else:
-        w["Intentional"] = w["Conventional"] = w["Arbitrary"] = round(1/3, 4)
-
-    # Triangle vertices
+        a_new = st.slider("Arbitrary", 0.0, 1.0, w["Arbitrary"], 0.01)
+
+    # --- Adjust other sliders proportionally ---
+    # Detect which slider changed
+    if i_new != prev_w["Intentional"]:
+        diff = i_new - prev_w["Intentional"]
+        total_other = w["Conventional"] + w["Arbitrary"]
+        if total_other > 0:
+            w["Conventional"] -= diff * (w["Conventional"] / total_other)
+            w["Arbitrary"] -= diff * (w["Arbitrary"] / total_other)
+        w["Intentional"] = i_new
+
+    elif c_new != prev_w["Conventional"]:
+        diff = c_new - prev_w["Conventional"]
+        total_other = w["Intentional"] + w["Arbitrary"]
+        if total_other > 0:
+            w["Intentional"] -= diff * (w["Intentional"] / total_other)
+            w["Arbitrary"] -= diff * (w["Arbitrary"] / total_other)
+        w["Conventional"] = c_new
+
+    elif a_new != prev_w["Arbitrary"]:
+        diff = a_new - prev_w["Arbitrary"]
+        total_other = w["Intentional"] + w["Conventional"]
+        if total_other > 0:
+            w["Intentional"] -= diff * (w["Intentional"] / total_other)
+            w["Conventional"] -= diff * (w["Conventional"] / total_other)
+        w["Arbitrary"] = a_new
+
+    # Clamp small floating point errors
+    for k in w:
+        w[k] = max(0.0, min(1.0, round(w[k], 4)))
+
+    # Update prev_weights for next run
+    st.session_state.prev_weights = w.copy()
+
+    # --- Triangle vertices ---
     vertices = np.array([
         [0.5, np.sqrt(3)/2],  # Intentional
         [0, 0],               # Conventional
@@ -56,7 +86,7 @@ def render():
         w["Arbitrary"] * vertices[2]
     )
 
-    # Plot
+    # --- Plot ---
     fig, ax = plt.subplots()
     ax.plot(*np.append(vertices, [vertices[0]], axis=0).T)
     ax.text(*vertices[0], "Intentional", ha="center", va="bottom", color="green", weight="heavy")
@@ -66,11 +96,10 @@ def render():
     ax.scatter(point[0], point[1], c="orange", s=10000, zorder=5, alpha=0.3)
     ax.set_aspect("equal")
     ax.axis("off")
-
     fig.patch.set_alpha(0)
     ax.patch.set_alpha(0)
 
-    # --- Dummy points scattered inside triangle ---
+    # --- Dummy points ---
     locations = [
         (0.9, 0.1, "Random Seeds", "Random Seeds are highly arbitrary, without any convention or intentionality.", "left", "bottom"),
         (0.35, 0.06, "Neural networks for Tabular Data", "Using neural networks of some arbitrary size (hidden layers) for a setting where they are not needed is highly conventional, a bit arbitrary, and has very low intentionality.", "left", "bottom"),
@@ -82,7 +111,6 @@ def render():
 
     torch_radius = 0.177
     explanations = []
-    # Illuminate nearby points
     for (x, y, label, labeltext, ha, va) in locations:
         dist = np.linalg.norm([x - point[0], y - point[1]])
         if dist <= torch_radius: