Upload MTGNN model trained on PEMS-BAY

README.md

@@ -21,9 +21,10 @@ This model uses a graph neural network architecture that combines:
 
 ## Evaluation Metrics
 
-- **MAE (test split)**: 1.7949
-- **MAPE (test split)**: 0.0444
-- **RMSE (test split)**: 3.5008
+- **Validation MAE**: 1.5766
+- **Test MAE**: 31.9022
+- **Test MAPE**: inf
+- **Test RMSE**: 44.5574
 
 
 ## Dataset
@@ -33,13 +34,15 @@ This model uses a graph neural network architecture that combines:
 ## Usage
 
 ```python
-from utils.mtgnn import load_from_hub, get_model_and_data
+from utils.mtgnn import load_from_hub
 
-# Download checkpoint
-checkpoint_path = load_from_hub("your-username/mtgnn-pems-bay")
+# Load model from Hub
+model = load_from_hub("PEMS-BAY")
 
-# Load model (requires dataset and config)
-# See documentation for full loading example
+# Get predictions
+import numpy as np
+x = np.random.randn(10, 2, 207, 12)  # (batch, features, nodes, seq_len)
+predictions = model.predict(x)
 ```
 
 ## Training
@@ -50,7 +53,7 @@ Model was trained using the MTGNN implementation with default hyperparameters.
 
 If you use this model, please cite the original MTGNN paper:
 
-```
+```bibtex
 @inproceedings{wu2020connecting,
   title={Connecting the Dots: Multivariate Time Series Forecasting with Graph Neural Networks},
   author={Wu, Zonghan and Pan, Shirui and Long, Guodong and Jiang, Jing and Chang, Xiaojun and Zhang, Chengqi},

best_model.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f606e09fc7b56bdf201dececffac393dc0819b0eaf0c2c2c6016d698eb5d5093
+oid sha256:52615ac439b7d032295df2f186026c4d35506ab385a1ee4864aa0cedea4b4e88
 size 2326709

config.json

@@ -1,6 +1,5 @@
 {
-  "gcn_true": true,
-  "buildA_true": true,
+  "num_nodes": 325,
   "gcn_depth": 2,
   "dropout": 0.3,
   "subgraph_size": 20,
@@ -10,19 +9,26 @@
   "residual_channels": 32,
   "skip_channels": 64,
   "end_channels": 128,
-  "seq_length": 12,
   "in_dim": 2,
-  "out_dim": 12,
+  "seq_in_len": 12,
+  "seq_out_len": 12,
   "layers": 3,
   "propalpha": 0.05,
-  "tanhalpha": 3,
-  "layer_norm_affline": true,
+  "tanhalpha": 3.0,
+  "batch_size": 64,
+  "val_batch_size": 8,
+  "epochs": 100,
   "learning_rate": 0.001,
   "weight_decay": 0.0001,
   "clip": 5,
-  "step_size": 100,
-  "epochs": 100,
-  "batch_size": 64,
-  "save_iter": 1,
-  "num_nodes": 325
+  "step_size1": 2500,
+  "step_size2": 100,
+  "gcn_true": true,
+  "buildA_true": true,
+  "cl": true,
+  "device": "cuda",
+  "seed": 101,
+  "print_every": 999999,
+  "save": "models/mtgnn_pems-bay/",
+  "dataset_name": "PEMS-BAY"
 }

metadata.json

@@ -1,10 +1,11 @@
 {
   "dataset": "PEMS-BAY",
-  "upload_date": "2025-11-05T20:28:34.444690",
+  "upload_date": "2025-11-07T19:05:10.987707",
   "metrics": {
-    "MAE (test split)": 1.7948928273018496,
-    "MAPE (test split)": 0.04443478034119363,
-    "RMSE (test split)": 3.5007667902075212
+    "Validation MAE": 1.5765657424926758,
+    "Test MAE": 31.90218734741211,
+    "Test MAPE": Infinity,
+    "Test RMSE": 44.55735397338867
   },
   "framework": "PyTorch",
   "model_type": "MTGNN"

model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:684a3a01da0d8ae181ea69bb969637bea3f025a72bcd2287316e55e712062be9
+size 2755777