Luciano Paz

I agree that the paper's method is a great posterior predictive check in itself. My proposal was much more mundane. Regarding the problem about the kde lines not being in...

@rrkarim, it would be great to add more samplers! We should think of a way to easily get the other tfp transition kernels

You can also run the tests locally by running `make test` in the project's root directory

Reopening because #238 only moved the fixtures from test sampling to conftest.py, but the rest of the test scripts have fixtures too. They should all be moved and potentially refactored...

@worasom, although it will sound strange, the problem you are seeing is caused because you installed the tensorflow package. I'll explain why. When we started developing pymc4, we made the...

I wouldn't add them as `Deterministics`. The executors are used in different contexts for different purposes. The transformed executor is used in `sample` and in `sample_posterior_predictive`, and the latter relies...

I can't check it out now, but the error seems to indicate there are some extra parameters that are being included in the initial value for the step. 9 instead...

Thanks for reporting this @jdehning. I can reproduce the exception you are running into and can confirm that its coming from the gradient computation. I still have to investigate more...

@gaow, @Sayam753, the error that is being raised comes from tensorflow's type promotion rules. The problem is that an int32 can't be promoted to a float32, because float32's can't represent...

@gaow, I hadn't seen the updated error. You are getting a gradient is None because `xi` is a discrete variable and it cannot be differentiated. Currently, pymc4 only performs NUTS...