Jeremie Coullon

Glad that my [SGMCMCJax](https://github.com/jeremiecoullon/SGMCMCJax) library might be helpful here! A tutorial to convert a NumPyro model to a logprior and a loglikelihood format would be great I agree. This is...

This is a good idea! I'll fix this soon.

Hi @zfurman56, sorry for never replying to this I was on holiday at the time and then forgot to get back to it 😓 Thanks very much for this PR...

Just merged your PR into master; thanks very much for this fix!

This is a great idea! 
This is related to [this issue](https://github.com/pyro-ppl/numpyro/issues/950#issuecomment-927118667) in NumPyro. However the thing to figure out here is how to convert PyMC3's log-posterior into a log-prior and...

@ricardoV94 : the log-prior graph sounds like the correct thing yes. However the log-likelihood would need to be a function without data "baked in". So a function like `log_likelihood(params, data)`....

> How do you make sure data is partitioned correctly? I don't quite understand what this means; could you explain some more please? In your example: `mu.shape == (N,)`, is...

Splitting data in the likelihood: the approach is exactly the same as in stochastic gradient descent (and related algorithms like Adam, RMSProp etc..). > For instance if you have a...

I added a [notebook](https://sgmcmcjax.readthedocs.io/en/latest/nbs/PyMC_integration.html) with a basic Gaussian model example working. Some questions: - In the PyMC sampling cell (cell number 5), the sampler takes much longer than the progress...

Thanks for getting touch! This is a very interesting idea; I would love to discuss this with you offline! I'll email you.