Christopher Suter

Thanks, both. What is the benefit to having a source tarball included? I can easily find the installed sources in, say, a virtualenv lib directory. Does this help some other...

No good reason this wasn't included. The existing tools were an intern project, so basically just what we could get done in time. We'd welcome a PR supporting more general...

thanks for sending. please add some unit tests covering the new behavior, in the corresponding test file. would any other layers benefit from adding the same new behavior?

Adding tf.function(...) wrapper to compile the code to XLA speeds up the cholesky_update for this problem size from the 1.38s you saw to around 40-60ms on my machine (nothing fancy)....

To implement the jax substrate, we basically reimplement the TF API surface using jax under the hood. This lets us leave most TFP code intact, just swapping out imports. This...

Wow, yeah, that's a pretty egregious bug. Thank you for flagging. There's pretty good test coverage of this distribution, including, eg, statistical tests of mean and variance, even in the...

Sorry for the sluggish response. Can you try this? ``` parameter_samples, _ = tf.function( func=lambda ots: tfp.sts.fit_with_hmc(model, ots), jit_compile=True, autograph=False)(observed_time_series) ```

There will be some compilation overhead on the first run but should be overall much faster.

What if you increase the quadrature_size parameter? The default of 8 may be too coarse. On Wed, May 13, 2020 at 10:37 pekaalto wrote: > Also checking: > > d...

@jvdillon @langmore anything jump out at you here?