Bytewise classifier training algorithm does not follow reference paper

[WojciechMigda]

From "Extending the Tsetlin Machine With Integer-Weighted Clauses for Increased Interpretability" arXiv:2005.05131:

Quote: As seen in the figure, if the clause output is a true positive, we simple increase the weight by 1. Conversely, if the clause output is a false positive, we decrease the weight by 1.

Tsetlini's bitwise algorithm does it correctly, whilst the bytewise algorithm decrements weights irrespective of clause output.

[WojciechMigda]

Reference implementation code (https://github.com/cair/pyTsetlinMachine): weights decrementation: https://github.com/cair/pyTsetlinMachine/blob/master/pyTsetlinMachine/ConvolutionalTsetlinMachine.c#L320 weights' incrementation: https://github.com/cair/pyTsetlinMachine/blob/master/pyTsetlinMachine/ConvolutionalTsetlinMachine.c#L339