Naweed Aghmad Khan

`model.train(losses)` losses assumes that many losses are given, i.e. a list of losses. `losses=[Loss.SUPERVISED]` solves this issue, but we should add an additional kwarg `loss` that only accepts a single...

We can start with byte-code, having the ability to dump and load a model is more important than having the dump human-readable right now.

> Removing quotation marks from `Smokes` helped (at least did not throw an error), but not sure what I am doing. Looks like there is some missing exception handling here....

Nice catch! Looks like the negation was incorrectly making assumptions about the ordering by using the default `None`. Unfortunately I can't merge in these changes without you signing off on...

Looks like there was some nuance that the None was encoding, that the table_rows are not. Can you please post the results for your 2 experiments, the problem may be...

Thanks for bringing this to our attention. @KyleErwin I've tested the first bit of code above, and it seems that #71 has even slowed down inference time of the original...

@asnota Can you retry the install again? The dependencies have been updated to remove some of the problematic libraries like graphviz/libgmp

#71 included the equivalence checks for objects to allow neurons and symbols to be matched. Since each model is simply a collection of connected formulae, we should have all objects...

since I don't have your model and can't recreate these results, I'm closing off the PR