Srinivas Vasudevan

Hi all, One of our internal 20%-ers Anudhyan has just added broadcasting support in the matmul (thanks again for the awesome work!): https://github.com/tensorflow/tensorflow/commit/47ab68d265a96b6e7be06afd1b4b47e0114c0ee9 What this will surface as a change...

The date for compatibility was delayed (to today). I am in the process of submitting a PR to make tf.linalg.LinearOperator use this new broadcasted op (so I expect people can...

Hi all, I recently added broadcasting triangular solve to TF, and switched LinearOperator to use this.

Broadcasting triangular solve: https://github.com/tensorflow/tensorflow/commit/b105944eb6c563849a085a1765d6700ee2c0f35c Using this in LinearOperator and a few other places: https://github.com/tensorflow/tensorflow/commit/0897278b482e7f015ac3cbc98b60eb0ed0f9bede

Yep that sounds good. Let me know if you see performance / memory issues. I can try to see if I can improve on this (I'll also some point in...

FWIW, we also do have a SoftmaxCentered bijector, which also constrains/unconstrains the same regions (R^n -> unit simplex in R^{n+1}), but can lead to different optimization landscapes / HMC trajectories...

Ordered should be here: https://www.tensorflow.org/probability/api_docs/python/tfp/bijectors/Ordered For Cholesky, I'm not sure what is exactly going on here, but we do have https://www.tensorflow.org/probability/api_docs/python/tfp/bijectors/TransformDiagonal which should help. Circular is something we are missing,...

This should be true for (0, 1) -> S^1 (or more generally for an interval (a, b), there's an injective map to S^1, which scales (a, b) and then applies...

1) We have `tfp.math.erfcx`, but that's meant for real arguments and probably not much code could be shared there. 2) Yep that sounds good. We have code for special functions...

1) Yeah separate and dispatching sounds fine. You can use https://cs.opensource.google/tensorflow/probability/+/master:tensorflow_probability/python/internal/dtype_util.py;l=145 to determine the complex dtype (this is mainly to be backend independent, so that this will work in JAX)....