Thanks for sharing this wonderful code.
One weird thing I've noticed is that the Persformer in /examples/orbit_5k_pipeline.ipynb is not trainable.
I have changed the Persformer hyperparameters and increased the dataset size, but the training accuracy remains around 20%, which is similar to random guessing.
I suspect there might be an issue with the Persformer code.
Maybe "Residual connections" described in Section 4.1 is missing in the code.
The paper reports that without the "Residual connections", the model was not trainable.
Describe the bug
Thanks for sharing this wonderful code. One weird thing I've noticed is that the Persformer in /examples/orbit_5k_pipeline.ipynb is not trainable. I have changed the Persformer hyperparameters and increased the dataset size, but the training accuracy remains around 20%, which is similar to random guessing.
I suspect there might be an issue with the Persformer code.