ICML revisions

Author: pchlenski

manify/predictors/kappa_gcn.py

@@ -417,6 +417,11 @@ def fit(
                 my_tqdm.update(1)
                 my_tqdm.set_description(f"Epoch {i+1}/{epochs}, Loss: {loss.item():.4f}")
 
+            # Early termination for nan loss
+            if torch.isnan(loss):
+                print("Loss is NaN, stopping training.")
+                break
+
         if use_tqdm:
             my_tqdm.close()
 

notebooks/26_benchmark_single_curvature_gaussian.ipynb

@@ -12,7 +12,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": 2,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -25,7 +25,7 @@
     "import matplotlib\n",
     "import matplotlib.cm\n",
     "\n",
-    "from embedders.predictors.tree_new import _angular_greater\n",
+    "from manify.predictors.decision_tree import _angular_greater\n",
     "\n",
     "######################\n",
     "# HELPER FUNCTIONS\n",
@@ -316,10 +316,9 @@
     "        color = class2color[cl]  # consistent across the entire tree\n",
     "        idx = ymask == cl\n",
     "        fig.add_trace(\n",
-    "        #     go.Scatter(x=x_in[idx], y=y_in[idx], mode=\"markers\", marker=dict(color=color), name=f\"Class {cl}\")\n",
+    "            #     go.Scatter(x=x_in[idx], y=y_in[idx], mode=\"markers\", marker=dict(color=color), name=f\"Class {cl}\")\n",
     "            go.Scatter(x=y_in[idx], y=x_in[idx], mode=\"markers\", marker=dict(color=color), name=f\"Class {cl}\")\n",
     "        )\n",
-    "        \n",
     "\n",
     "    # d) Finally set the axis range to the bounding box of the data (so we don't zoom out).\n",
     "    x_min, x_max = x_all.min(), x_all.max()\n",
@@ -343,7 +342,6 @@
     "######################\n",
     "\n",
     "\n",
-    "\n",
     "def create_dashboard(pdt, X, y):\n",
     "    node2id, edges = build_edges_and_id_map(pdt)\n",
     "    node_masks = compute_node_masks(pdt, X, y, node2id)\n",
@@ -403,9 +401,7 @@
     "            fig = go.Figure()\n",
     "            y_in_leaf = y[mask]\n",
     "            if len(y_in_leaf) == 0:\n",
-    "                fig.add_annotation(\n",
-    "                    x=0.5, y=0.5, xref=\"paper\", yref=\"paper\", text=\"Empty Leaf Node\", showarrow=False\n",
-    "                )\n",
+    "                fig.add_annotation(x=0.5, y=0.5, xref=\"paper\", yref=\"paper\", text=\"Empty Leaf Node\", showarrow=False)\n",
     "            else:\n",
     "                counts = Counter(y_in_leaf.tolist())\n",
     "                labels = list(counts.keys())\n",
@@ -431,24 +427,22 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 21,
+   "execution_count": null,
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "0.6250\n"
+      "0.7050\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/var/folders/ck/0ybgtq694jnd4mbjw_0rm6dh0000gp/T/ipykernel_21314/3256736440.py:153: MatplotlibDeprecationWarning:\n",
-      "\n",
-      "The get_cmap function was deprecated in Matplotlib 3.7 and will be removed two minor releases later. Use ``matplotlib.colormaps[name]`` or ``matplotlib.colormaps.get_cmap(obj)`` instead.\n",
-      "\n"
+      "/tmp/ipykernel_89533/2928454690.py:153: MatplotlibDeprecationWarning: The get_cmap function was deprecated in Matplotlib 3.7 and will be removed two minor releases later. Use ``matplotlib.colormaps[name]`` or ``matplotlib.colormaps.get_cmap(obj)`` instead.\n",
+      "  tab10 = matplotlib.cm.get_cmap(\"tab10\")\n"
      ]
     },
     {
@@ -466,27 +460,27 @@
        "        "
       ],
       "text/plain": [
-       "<IPython.lib.display.IFrame at 0x36881dd80>"
+       "<IPython.lib.display.IFrame at 0x757bbfd23670>"
       ]
      },
      "metadata": {},
      "output_type": "display_data"
     }
    ],
    "source": [
-    "import embedders\n",
+    "import manify\n",
     "from sklearn.model_selection import train_test_split\n",
     "\n",
-    "pm = embedders.manifolds.ProductManifold(signature=[(1, 2)])\n",
-    "X, y = embedders.gaussian_mixture.gaussian_mixture(pm, 1000, num_classes=4)\n",
+    "pm = manify.ProductManifold(signature=[(1, 2)])\n",
+    "X, y = pm.gaussian_mixture(1000, num_classes=4)\n",
     "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)\n",
     "\n",
-    "pdt = embedders.predictors.tree_new.ProductSpaceDT(pm=pm, n_features=\"d_choose_2\", max_depth=5)\n",
+    "pdt = manify.ProductSpaceDT(pm=pm, n_features=\"d_choose_2\", max_depth=5)\n",
     "pdt.fit(X_train, y_train)\n",
     "\n",
     "print(f\"{pdt.score(X_test, y_test).float().mean().item():.4f}\")\n",
     "\n",
-    "create_dashboard(pdt, X_test, y_test).run_server(debug=True)"
+    "create_dashboard(pdt, X_test, y_test).run(debug=True)"
    ]
   },
   {
@@ -499,7 +493,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "base",
+   "display_name": "manify",
    "language": "python",
    "name": "python3"
   },
@@ -513,7 +507,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.14"
+   "version": "3.10.0"
   }
  },
  "nbformat": 4,