Brian Patton
Brian Patton
It should be super(SkewNormal, self).foo Curious: why do you also subclass Normal instead of just Distribution? On Fri, Sep 6, 2019, 6:09 AM inhandan wrote: > No worries! Thanks for...
The call to super should generally have the class in which the method is implemented as its first argument. This is pretty conventional python (though in py3 you just write...
Are you ready for us to take another look at this? Just ping when you think it's ready.
@csuter recently added some helper scripts to make testing easier https://github.com/tensorflow/probability/blob/master/CONTRIBUTING.md#helper-scripts
To your immediate question about tests, the answer is yes, you should add some basic tests in a corresponding test.py file. You can also get some good test coverage by...
If the distribution has a tractable cdf I think using the DKWM bound test should suffice to validate sample and cdf consistency. Gamma has an example here. https://cs.opensource.google/tensorflow/probability/+/master:tensorflow_probability/python/distributions/gamma_test.py;l=418 To validate...
Is this ready to review? At a glance the tests look OK.
Actually MixtureSameFamily doesn't implement quantile. We could do that the way you suggest. PRs welcome.
Oh never mind, was thinking backward. Cdf you can complete with weighted sum, and is already implemented. Not quantile.
There is now support for sobol sequences in https://www.tensorflow.org/api_docs/python/tf/math/sobol_sample