Utkarsh

@dbtsai Do you have some kind of performance benchmarks for it. It would be really of great help .

@ankurankan @npfp Even I'm not sure about any benefit of doing so. But I don't think pre-count would be available in most of the cases.

@Harshvardhan1 as @yashu-seth mentioned you need to write tests for whatever change you do. Read our contribution guidelines and [this section](https://github.com/pgmpy/pgmpy/blob/dev/Contributing.md#tests). You can look at tests for MakovChain at [here](https://github.com/pgmpy/pgmpy/blob/dev/pgmpy/tests/test_models/test_MarkovChain.py#L110)...

@MadDataScience > I noticed that, when given multiple variables, _variable_elimination computes > them each independently, marginalizing over the others. This is a problem when > the variables are not independent....

@ankurankan @yujianll I think exception might be coming because the BIFReader loads the whole file in-memory.

@ankurankan Lemme try this with pytorch. I'm thinking we should have alternate computation backends like pymc.

@ankurankan When I was working on HMC and NUTS I couldn't find any papers of where these were being used in case of Bayesian Networks. But yeah, mathematically we should...

@JustinTervala Please rebase your branch to latest dev. You can read about rebase [here](https://www.atlassian.com/git/tutorials/merging-vs-rebasing). Also, please squash the commits into a single commit.

@ankurankan @JustinTervala I believe we should also add an example on how to create GaussianDistribution using string.

@wqp89324 Can you share the code and relevant data (if any)? We cannot understand the cause of error without the code.