pyconstruct.learners.BlockCoordinateFrankWolfe

class pyconstruct.learners.BlockCoordinateFrankWolfe(domain=None, model=None, structured_loss=None, dataset_size=1, alpha=0.0001, **kwargs)

Learner using the Block-Coordinate Frank-Wolfe algorithm [1].

This implementation is still a bit experimental. Should work fine but still has not been made parallel, so it will take much more than necessary on most datasets. Coming soon also improvements from [2].

Parameters:

• domain (BaseDomain) – The domain of the data.
• structured_loss (function (y, y) -> float) – The structured loss to compute on the objects.
• dataset_size (int) – A hint on the size of the dataset. May improve performance.
• alpha (float) – The regularization coefficient.

References

[1]	Lacoste-Julien, Simon, et al. “Block-Coordinate Frank-Wolfe Optimization for Structural SVMs.” ICML 2013 International Conference on Machine Learning. 2013.

[2]	Osokin, Anton, et al. “Minding the Gaps for Block Frank-Wolfe Optimization of Structured SVMs.” Proceedings of Machine Learning Research. Proceedings of the International Conference on Machine Learning (ICML 2016). 2016.

Methods

decision_function(X, Y, **kwargs)
fit(X, Y[, Y_pred])	Updates the current model with a mini-batch (X, Y).
get_params([deep])	Get parameters for this estimator.
loss(X, Y_true, Y_pred)
partial_fit(X, Y[, Y_pred])	Updates the current model with a mini-batch (X, Y).
phi(X, Y, **kwargs)	Computes the feature vector for the given input and output objects.
predict(X, *args, **kwargs)	Computes the prediction of the current model for the given input.
score(X, Y[, Y_pred])	Compute the score as the average loss over the examples.
set_params(**params)	Set the parameters of this estimator.

Attributes

dual_gap