SparseGAP implementation #76
Conversation
Co-authored-by: Philip Loche <[email protected]>
Co-authored-by: Philip Loche <[email protected]>
* Write train output to hydra's output directory * Added evaluation function * Add usage example for cli interface * update train cli * Disable MacOS tests * Add cli skeleton for exporter --------- Co-authored-by: frostedoyster <[email protected]>
* Add gradient calculator * Temporary losses * Forces and stresses * Support multiple model outputs in SOAP-BPNN
.github/workflows/gap-tests.yml
Outdated
|
|
||
| jobs: | ||
| tests: | ||
| runs-on: ${{ matrix.os }} |
There was a problem hiding this comment.
I think we don't need a Matrix here. We have just ubuntu.
src/metatensor/models/experimental/gap/tests/test_regression.py
Outdated
Show resolved
Hide resolved
| ] | ||
|
|
||
| __maintainers__ = [ | ||
| ("Davide Tisi <[email protected]>", "@DavideTisi"), |
There was a problem hiding this comment.
We'll need to add @DavideTisi to the CODEOWNERS files as well
| print(ref_output["mtm::U0"].block().values) | ||
| print(scripted_output["mtm::U0"].block().values) |
There was a problem hiding this comment.
| print(ref_output["mtm::U0"].block().values) | |
| print(scripted_output["mtm::U0"].block().values) |
| logger = logging.getLogger(__name__) | ||
|
|
||
|
|
||
| class Trainer: |
There was a problem hiding this comment.
I'm not sure I follow what's going on here. This seems to already be using #185 which is not merged yet.
--------- Co-authored-by: frostedoyster <[email protected]>
|
Note: need better error message when number of sparse points is greater than the number of atomic environments in the training set |
|
Note: we may want to save GAP checkpoints |
|
Note: we need to fix eval which has the old |
if n_sparse_point > environments
what's a gap checkpoints? |
if the number of sparse env is bigger than the number of environments
For GAP, the checkpoint is the same thing as the final model. But having checkpoints allow use to have the same interface as the other architectures, especially for final export. |
A draft of SparseGAP that can predict energies. It contains the code of PR lab-cosmo/equisolve#61. The utilities used in this PR will be merged to corresponding
metatensor-{operations,learn}modules and then replaced here, but for the sake of having something working that can be tried out the current version is self-contained.TODOs
📚 Documentation preview 📚: https://metatensor-models--76.org.readthedocs.build/en/76/