Forward pass in tutorial notebook

pyt-team · Jul 13, 2024 · 5a91883 · 5a91883
1 parent 0b221bb
commit 5a91883
Show file tree

Hide file tree

Showing 2 changed files with 39 additions and 3 deletions.
diff --git a/configs/datasets/wall_shear_stress.yaml b/configs/datasets/wall_shear_stress.yaml
@@ -2,3 +2,12 @@ data_domain: simplicial_complex
 data_type: artery
 data_name: wall_shear_stress
 data_dir: datasets/${data_domain}/${data_type}
+
+num_features:
+  - 1 # initial node features
+  - 3 # initial edge features
+num_classes: 3
+task: regression
+loss_type: mae
+montitor_metric: mae
+task_level: node
diff --git a/tutorials/pointcloud2hypergraph/pointnet_lifting.ipynb b/tutorials/pointcloud2hypergraph/pointnet_lifting.ipynb
@@ -33,6 +33,7 @@
     "from modules.utils.utils import (\n",
     "    describe_data,\n",
     "    load_dataset_config,\n",
+    "    load_model_config,\n",
     "    load_transform_config,\n",
     ")"
    ]
@@ -94,7 +95,8 @@
     "    pos=dataset.pos[slice_pos],\n",
     "    face=dataset.face[:, slice_face],\n",
     "    # x_2=dataset.x_2[slice_face],\n",
-    "    # incidence_2=dataset.incidence_2[slice_pos, slice_face]  # not supported by PyTorch\n",
+    "    # incidence_2=dataset.incidence_2[slice_pos, slice_face]  # not supported by PyTorch,\n",
+    "    num_features=dataset.num_features,\n",
     ")\n",
     "\n",
     "print(\"Data sample:\")\n",
@@ -122,10 +124,10 @@
     "transform_config[\"lifting\"][\"sampling_ratio\"] = 0.2\n",
     "transform_config[\"lifting\"][\"cluster_radius\"] = 0.1\n",
     "\n",
-    "lifted_dataset = PreProcessor(\n",
+    "lifted_data = PreProcessor(\n",
     "    data, transform_config, os.path.join(rootutils.find_root(), dataset_config.data_dir)\n",
     ")\n",
-    "describe_data(lifted_dataset)"
+    "describe_data(lifted_data)"
    ]
   },
   {
@@ -134,6 +136,31 @@
    "source": [
     "This hypergraph represents the first set abstraction layer that is used by [PointNet++](https://arxiv.org/abs/1706.02413). To construct a complet PointNet++ out of this, we would have to recursively apply the lifting while regarding the previous hyperedges as new \"hyper-nodes\"."
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Running inference"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from modules.models.hypergraph.unigcn import UniGCNModel\n",
+    "\n",
+    "model_type = \"hypergraph\"\n",
+    "model_id = \"unigcn\"\n",
+    "model_config = load_model_config(model_type, model_id)\n",
+    "\n",
+    "model = UniGCNModel(model_config, dataset_config)\n",
+    "\n",
+    "print(\"\\nModel output:\")\n",
+    "model(lifted_data)"
+   ]
   }
  ],
  "metadata": {