skada
skada copied to clipboard
scikit-adaptation
Domain adaptation toolbox compatible with scikit-learn and pytorch
Changed metrics.py to add a new parameter 'criterion' within 'loss_function' with default as 'cross_entropy_loss' and can switch to 'BCELoss'
The function to create deep DA models should have a parameter `criterion` or `base_criterion` or `source_criterion`. For now, it is hardcoded as a cross-entropy loss but if the model has...
Error raised: `TypeError: Singleton array array(None, dtype=object) cannot be considered a valid collection.` To reproduce: ``` from sklearn.linear_model import LogisticRegression from sklearn.model_selection import ShuffleSplit, cross_validate from skada import make_da_pipeline, MMDTarSReweightAdapter...
By using `SelectTarget`, the estimator only sees the target samples/labels/domains. However to compute the `CircularValidation` we need to train the estimator on the source THEN the target samples. Thus `ValueError:...
While this feature makes no sens for fitting, it can be very practical for multi-source/target for instance it would allow to sometging like ```python pipe = make_da_pipeline( PerDomain(StandardScaler()), SelectSource(SVC()), )...
 Hi, I played a bit with the logo. Let me know what you think about it. Maybe I will still change a bit the colors
Right now both are excluded from processing in `ruff.toml`.
This is required for us to understand how the API "feels" when working with multi-source multi-target settings. It's also crucial for properly covering the functionality of selectors (note that as...
We need to provide way better example of the deep methods. For each methods we should provide a training example (which is already done) and then an explanation of what...
Deep methods do not currently support `device='gpu'`. It should break at several locations when we define a new tensor or a `domain_classifier` for the adversarial methods. This issue should not...
Metadata
Owner
Metadata
Domain adaptation toolbox compatible with scikit-learn and pytorch