Merge pull request #28 from loicbachelot/main

[update] finish cleanup for geolab

Author: loicbachelot

_toc.yml

@@ -11,7 +11,6 @@ parts:
 - caption: Seismic Phase Detection
   chapters:
   - file: notebooks/niyiyu/curated_pnw_dataset_seisbench.ipynb
-  - file: notebooks/Loic/UNet/add_training_split.ipynb
   - file: notebooks/Loic/UNet/make_unet_annotated.ipynb
   - file: notebooks/Amanda/unet_performance_metrics.ipynb
   - file: notebooks/Amanda/apply_unet.ipynb

notebooks/Loic/GNN/GNN_node_regression_annotated.ipynb

@@ -5,7 +5,7 @@
    "id": "7a56d467",
    "metadata": {},
    "source": [
-    "# 🧠 GNN for Node-Level Regression: Predicting Sine Waves\n",
+    "# 🌐 GNN for Node-Level Regression: Predicting Sine Waves\n",
     "\n",
     "**Author:** Loïc Bachelot  \n",
     "**Goal:** This notebook demonstrates how to use a GNN to predict 1D sine wave signals at each node of a synthetic graph.\n",
@@ -491,7 +491,7 @@
     "- **`signal`**: Node signals (sine waves delayed by distance to a hidden origin) `[num_nodes, signal_length]`\n",
     "- **`origin`**: The hidden 2D origin point `[2]` used to generate node signals (not directly used for supervision here)\n",
     "- **`edge_index`**: Connectivity information between nodes (edges)\n",
-    "- **`edge_weight`**: Inverse-distance weights for edges (higher weight = closer nodes)\n",
+    "- **`edge_attr`**: distance weights for edges (both in x and y)\n",
     "\n",
     "This compact graph-based representation allows flexible training for node-level prediction tasks,  \n",
     "while preserving spatial and relational structure essential for learning.\n"
@@ -632,7 +632,7 @@
     "### 🏗️ Architecture Overview:\n",
     "\n",
     "1. **Signal Feature Extraction**  \n",
-    "   Each node’s raw signal is processed through a 1D convolution and max pooling to extract compact local features.\n",
+    "   Each node’s raw signal is processed through a simple MLP to extract compact local features.\n",
     "\n",
     "2. **Feature Fusion**  \n",
     "   The extracted signal features are concatenated with the node's spatial coordinates (x, y) to create a combined feature vector.\n",