We need to implement Adaptive downsampling in transformertf in transformertf using for instance RDP. This should be implemented in the LightningDataModuleBase, and should take place after transforms are applied - i.e. after normalization of data.
Furthermore, relative time indices need to be added to the data, since RDP result in irregularly sampled time series. The relative time indices should also be normalized. Optionally, logarithmic transforms can be added prior to normalization if time gaps contain outliers.
- Implement adaptive downsampling in transformertf [priority:: medium] [completion:: 2024-08-06]
- Implement relative time indices in transformertf [priority:: medium] [completion:: 2024-07-23]
We don’t need to implement adaptive downsampling in transformertf. The user can simply perform the adaptive downsampling prior to fitting, since adaptive downsampling can be a resource-intensive process.