secml / secml.github.io Goto Github PK

On line 155 of the c_classifier file self._fit(CDataset(data_x, dataset.Y), n_jobs=n_jobs) continues to error, removing the second parameter resolves this, I believe this should be removed as it appears parallel processing is not implemented correctly

The function explain function in secml.explanation.c_explainer_influence_functions.CExplainerInfluenceFunctions should compute the Hessian matrix of the model's parameters with respect to the training points, see Koh, P. W., & Liang, P. (2017, July). Understanding black-box predictions via influence functions.

Indeed, the current implementation gets the Hessian with respect to the sample given at test time.

Line 88 should be:

H = self.hessian(self.tr_ds.X, self.tr_ds.Y)