The transformer architecture proposed in this work is inspired by the DecisionTransformer architecture implemented in the HuggingFace library [1].
Our implementation can be found in the
| Parameter description | Value |
|---|---|
| Embedding dimension | |
| Maximum context length | |
| Number of layers | |
| Number of attention heads | |
| Batch size | |
| Non-linearity | |
| Dropout | |
| Learning rate | |
| Learning rate decay | |
| Gradient norm clip | |
| Gradient accumulation iters |
[1] “Huggingface’s Tranformers Library”, https://huggingface.co/docs/transformers/index.
| Parameter description | Symbol | Value |
|---|---|---|
| Number of samples in the dataset | ||
| Number of REL solutions in the dataset | ||
| Number of SCP solutions in the dataset | ||
| Train split (%) | - | |
| Test split (%) | - |
| Parameter description | Symbol | Value |
|---|---|---|
| Interaction with the environment collected at each |
||
| Possible values for the planning horizon for each interaction | ||
| Initial open-loop to closed-loop ratio in the aggregated dataset | - | |
| Train split (%) | - | |
| Test split (%) | - |