Open kmike opened 10 years ago
The python wrapper loads any supplied model and syncs it before training, just like the c-binary, so there shouldn't be a problem. However, the python wrapper does not process any new patterns passed in if you also pass a model; could that be the problem in your case? Let me know if you find a reproducible example!
It seems there is some shared state that is not encapsulated in Model class: I recall model.train continues training where it left for previous model; sometimes it exits with "error: invalid feature: " error on second training (while working on first).
Sorry, no reproducable example to check that :)