support wrapped models in grid search

[andresliszt]

Hello!, thanks for this amazing project. I am experiencing the same issue as the one mentioned in here . This pull request aims to add a way to include the arguments of the wrapped model in the grid, instead of just passing a list of instances of the wrapped model to the grid constructor

Fixes #2104 .

Summary

When the model key is found in the parameters dictionary passed to the gridsearch classmethod (meaning we are in the context of a model that wraps another), the class method expects either a list of wrapped model instances , or it expects a dictionary with a special key called model_class whose value is the class of the model to be wrapped. The other keys in the dictionary are the parameters that will be used to construct the grid dedicated to the wrapped model. Example

from sklearn.ensemble import RandomForestRegressor

from darts.models import RegressionModel
from darts.utils import timeseries_generation as tg

parameters = {
    "model": {
        "model_class": RandomForestRegressor,
        "min_samples_split": [2,3],
        "min_samples_leaf": [1,2],
    },
    "lags": [1,2,3],
}
series = tg.sine_timeseries(length=100)

RegressionModel.gridsearch(
    parameters=parameters, series=series, forecast_horizon=1
)

[andresliszt]

Interesting the test darts/tests/models/forecasting/test_transformer_model.py::TestTransformerModel::test_fit is failing with :
FileNotFoundError: [Errno 2] No such file or directory: '/var/folders/0t/rk2xtsbd6ws0rz79rbxkjrv00000gr/T/dartse01ues80/unittest-model-transformer/checkpoints/last-epoch=1.ckpt'

In the master branch i get the same error, so i think something with torch save method might be broken and is not related with this PR

[dennisbader]

Hi @andresliszt, the failing unit tests come from PyTorch Lightning version 2.1.3.
The version changed saving the last checkpoint as a symlink which brakes our tests.
They already reverted this here and will be released in the next version.

For now I'll make a PR to set an an upper cap on pytorch_lightning<=2.1.2, and will relax it once they release the new version.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (ab3d77a) 93.92% compared to head (910bef1) 93.92%.
Report is 1 commits behind head on master.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2133   +/-   ##
=======================================
  Coverage   93.92%   93.92%           
=======================================
  Files         135      135           
  Lines       13335    13338    +3     
=======================================
+ Hits        12525    12528    +3     
  Misses        810      810           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.