Update app.py

app.py

@@ -17,7 +17,6 @@ def generate_hypercube(n=4):
     return np.array(list(itertools.product([0, 1], repeat=n)), dtype=float)
 
 def generate_simplex(n=3):
-    # n-simplex in n-D: standard basis vectors + origin
     eye = np.eye(n, dtype=float)
     origin = np.zeros((1, n), dtype=float)
     return np.vstack([eye, origin])
@@ -44,8 +43,7 @@ def parse_text_points(text: str) -> np.ndarray:
     txt = textwrap.dedent(text.strip())
     rows = [r for r in txt.splitlines() if r.strip()]
     data = [list(map(float, r.replace(",", " ").split())) for r in rows]
-    arr = np.array(data, dtype=float)
-    return arr
+    return np.array(data, dtype=float)
 
 def run_tsne(data, perp, seed):
     ts = TSNE(n_components=2, perplexity=perp, random_state=seed, init="pca")
@@ -95,13 +93,11 @@ with st.sidebar:
     algo = st.selectbox("Method", ["t-SNE","PCA","UMAP"])
     seed = st.number_input("Random seed", value=42, step=1)
 
-    # method-specific
     if algo == "t-SNE":
         perp = st.slider("Perplexity", 5.0, 50.0, 30.0, 1.0)
     elif algo == "UMAP":
         neighbors = st.slider("n_neighbors", 5, 200, 15, 5)
         min_dist = st.slider("min_dist", 0.0, 0.99, 0.1, 0.01)
-    # PCA has no extra params
 
     st.header("3️⃣ Clustering (optional)")
     do_cluster = st.checkbox("Cluster embedding")
@@ -130,10 +126,15 @@ if run:
     else:
         emb, kl = run_umap(pts, neighbors, min_dist, seed)
 
-    # compute trustworthiness
-    tw = trustworthiness(pts, emb, n_neighbors=5)
+    # dynamic trustworthiness
+    n_samples = pts.shape[0]
+    k_max = (n_samples - 1) // 2
+    if k_max >= 1:
+        tw = trustworthiness(pts, emb, n_neighbors=k_max)
+    else:
+        tw = None
 
-    # clustering
+    # clustering & plotting
     df = pd.DataFrame(emb, columns=["x","y"])
     if do_cluster:
         if cluster_algo == "KMeans":
@@ -154,7 +155,11 @@ if run:
     st.subheader("2-D Embedding")
     st.plotly_chart(fig, use_container_width=True)
 
-    st.markdown(f"**Trustworthiness (k=5):** {tw:.3f}")
+    if tw is not None:
+        st.markdown(f"**Trustworthiness (k={k_max}):** {tw:.3f}")
+    else:
+        st.markdown("**Trustworthiness:** Not enough samples to compute (need ≥3 points).")
+
     if kl is not None:
         st.markdown(f"**t-SNE KL divergence:** {kl:.3f}")
 
@@ -167,11 +172,9 @@ if run:
         mime="text/csv"
     )
 
-    # raw data expander
     with st.expander("Show original data"):
         st.write(pts)
 
-    # t-SNE math explainer
     if algo == "t-SNE":
         with st.expander("🧠 How t-SNE works"):
             st.markdown(r"""