This is an attempt to port to pomegranate 1.0. Current code is not working yet, even though most of the API is ported.
In the current state, the models don't seem to get fitted, as we get a negative improvement as in the first iteration.
It could be that there are still bugs in the library itself causing this
I am also not sure, if I understood/implemented the new concept of priors correctly.
Biggest issue is that we need to retrain the old model or find some hacky way to port it to the new format.
It seems like I made some progress, by fixing the initial probability matrix and tweaking how priors are calculated for the training. Now at least, we don't get negative improvements and the model seems to train.
This is an attempt to port to pomegranate 1.0. Current code is not working yet, even though most of the API is ported.
In the current state, the models don't seem to get fitted, as we get a negative improvement as in the first iteration. It could be that there are still bugs in the library itself causing this
I am also not sure, if I understood/implemented the new concept of priors correctly.
Biggest issue is that we need to retrain the old model or find some hacky way to port it to the new format.