jmschrei / torchegranate Goto Github PK

Hi there,

I work in a pycharm environment, and I have installed pomegranate 1.0 and torchegranate 0.5.0, and tried it with both libraries.
After installing the library it does not recognize any of the commands.

For example running this code:

from pomegranate import *

test = State(DiscreteDistribution({
    "good": 0.7,
    "bad": 0.2,
    "medium": 0.1
}), name="test")

returns the error:

NameError: name 'State' is not defined

The same happens when I replace pomegranate with torchegranate.

I use python 3.10.

Why could this be happening?

Is there a way to create a model using a univariate normal? I am forced to use at least 2 variables since:

linalg.cholesky: The input tensor A must have at least 2 dimensions.

in normal.py, l.139 chol = torch.linalg.cholesky(self.covs)

See below in readme, not sure what should be in its place

Thought it may be helpful/convenient to have a link to the license in the readme

Hi, long-time user of pomegranate here. Thanks for the great work over the years!

My question: Is there any plan to bring the IndependentComponentsDistribution from pomegranate into torchegranate? I really liked the option of modeling each dimension of the emission with a specific distribution/mixture model.

Of course if there's another way to do this in torchegranate that I'm just too dumb to realize, I'd also appreciate that info. :)