From 0.8.* to 0.9

Make sure you view this update guide from the tag (version) of SecML you would like to install. In most cases this should be the highest numbered production tag (without rc in it).

To update the current v0.8.* version to v0.9, run the following steps:

1. Configuration file

The following new configuration settings are now available. Please update your $SECML_HOME_DIR/secml.conf file if needed, otherwise default values will be used.

  1. Added new section [secml:pytorch] to control the behaviour of the classes related to pytorch library support.
    The following options are available:

    • data_dir. Directory for storing pytorch data. Default: SECML_HOME_DIR/pytorch-data.

    • use_cuda. True (default) if CUDA should be used by the pytorch wrapper.

2. Deprecations

The following classes, methods or functions are now deprecated.

  • CClassifierKDE: use CClassifierSkLearn with sklearn.neighbors.KernelDensity instead

  • CClassifierMCSLinear: use CClassifierSkLearn with sklearn.ensemble.BaggingClassifier instead

  • CPreProcess.revert(): use .inverse_transform() method instead