Hi @Davide-Miotti , thank you for your valuable contribution!

I would suggest integrating the entire TimeWeighting logic directly into the autoregressive solver. In PINA, weights are typically used to balance different conditions. In this case, however, the weights correspond to different time steps within the same condition, so placing this logic inside the weighting mechanism could be confusing and potentially misleading.

To give some additional context, here is a python script that uses the code from this PR:
autoregressive_prova.py

With the last commit I deleted the autoregressive condition and now all the logic is contained within autoregressive solver, I also pushed by mistake a test script, called autoregressive_prova_generic_condition.py which can be used to test the new pipeline. I'll live it there for you to try and remove it later before eventual merge.
Now the solver is set to use DataCondition, as was suggested by @dario-coscia and adopts an adaptive weighting for unrolls as suggested by @GiovanniCanali
Side note, the automatic random shuffling performed by the dataloader is disruptive for autoregressive method, in the test script I set it to false in the Trainer settings

Hi @GiovanniCanali @Davide-Miotti, since this PR is related to 0.3 release, I changed the base branch

With the last edits I moved the whole logic to AutoregressiveSolver in pina/solver/autoregressive_solver/autoregressive_solver.py and to AutoregressiveSolverInterface in pina/solver/autoregressive_solver/autoregressive_solver_interface.py. The logic for splitting input data into multiple unrolls is implemented through a @dataclass UnrollInstructions defined in the same file as AutoregressiveSolverInterface.

This solver does not support subdivision into training and validation data. I believe we should wait for latest version of conditions and data loader for that one, because any random splitting of the data breaks the temporal sequence required by the autoregressive solver.

I am preparing a test script that show how it all works, will load it in the next days
autoregressive_codice_prova.py

Implemented the changes and instead of preparing a separate script, I added an end-to-end test that shows how the solver can be used.
Also corrected the docstrings, do I need to also update the documentation somewhere else?

Implemented the changes and instead of preparing a separate script, I added an end-to-end test that shows how the solver can be used. Also corrected the docstrings, do I need to also update the documentation somewhere else?

Thank you @Davide-Miotti!

Yes, you also need to add the .rst files under PINA/docs/source/_rst. In particular, please update _code.rst accordingly and add a <file_name>.rst file for each new file you introduced. You can refer to the existing files in that directory as a reference.

Memo: move the unroll logic and the settings dictionary workaround to the AutoregressiveCondition once available.

Hey! Very good job! Since I see many commit messages can you squash in two/three main commits?

Hey! Very good job! Since I see many commit messages can you squash in two/three main commits?

Hi @dario-coscia,
We will squash the commits before merging. Since this is still work in progress, we prefer to keep the history as is for now to avoid losing any relevant changes.