OSCAR4 for Debian ----------------- The pre-trained models (NGram models and DFAONTCPRFinder) have been excluded to comply with DFSG and Policy for Debian & Machine Learning [1]. According to [2], > The serialised model inherited from OSCAR3 was customised using > unknown training data and due to a serialisation bug is missing > probabilities at the end of the data array. This method is currently > retained for convenience during testing. Instead of ChEBI ontology included in the tarball, the one installed by emboss-data Debian package is used. This enables using most recent ChEBI ontology as shipped by emboss-data. [1] https://salsa.debian.org/deeplearning-team/ml-policy/-/tags/v0.2.5 [2] oscar4-recogniser-core/src/main/java/uk/ac/cam/ch/wwmm/oscarrecogniser/tokenanalysis/NGram.java -- Andrius Merkys Thu, 08 Sep 2022 03:03:07 -0400