cambridge-mlg / einsumnetworks Goto Github PK

I have seen that ARt-SPN can be classified, and I plan to conduct the same classification in EinsumNetworks, but the training accuracy is very low, I can't find the reason, can you add its classification effect? I look forward to receiving your reply. Thank you.

By setting use_em=False (e.g., to use SGD), the reparam function is created as a local function which cannot be readily pickled.

This means that torch.load and torch.save (which are using pickle) will throw exceptions like:

AttributeError: Can't pickle local object 'SumLayer.reparam_function.<locals>.reparam'

Using load_from_state seems not to be supported out of the box.
To reproduce this behaviour, see this minimal working example https://github.com/arranger1044/EinsumNetworks-1/blob/master/test/test_load_save.py

Hi,
I would like to know if we can specify different exponential families as leaf distributions for different random variables. This would be helpful in modelling tabular data, which typically comprises of both continuous and discrete valued features. Thanks.