Benjamin Bossan
Benjamin Bossan
#### Reference Issues/PRs #22893 #### What does this implement/fix? Explain your changes. This PR adds metadata routing to `CalibratedClassifierCV` (CCV). CCV uses a subestimator to create (out of sample) probabilities,...
~This is WIP, more documentation and tests are still missing. This draft should serve to discuss the design.~ _Update_ ## Description This PR adds the possibility to create model checkpoints...
Supersedes #877 ## Description This bug could occur if a user has set a parameter with a CUDA dependency and then tries to load the net without CUDA. Now, this...
Solves #848 ## Description For `Checkpoint`, as of now, `dirname` can only be a string. With this update, it can also be a callable with no arguments that returns a...
Fixes #718 [This notebook](https://github.com/skorch-dev/skorch/blob/feature/optuna-example-718/notebooks/optuna-example.ipynb) shows how to use skorch in conjunction with optuna, which can be much more efficient than `GridSearchCV` or `RandomizedSearchCV`. I don't have a deep knowledge of...
It would be nice to add an example of using [optuna](https://github.com/optuna/optuna) with skorch. The main reason is that `GridSearchCV` and `RandomizedSearchCV` can be much too slow for certain neural networks....
Currently, we don't store the learning rate but it is now being exposed in PyTorch (we may need to check the versions though). Reported by @jiajiexiao [here](https://github.com/skorch-dev/skorch/pull/738#discussion_r568267734).
Training with Cuda on a GPU machine, it is not possible to load the model on a machine without GPU in a straightforward fashion because some theano parameters are Cuda...
Only basic implementation yet, not finished. Idea: When wrapping dict input in a SliceDict, we could hopefully the reduce number of `if isinstance(X, dict) ... else ...` but probably not...
Scores based on `np.average` are wrong when input is a dictionary. The reason is that `len` is called on the batch, which returns the number of keys on a dict....