PyStan vs TFP

[Sarthak Kala]

Tried running a regression model with both TFP and PyStan. PyStan is atleast 10x faster. Is there a reason why? Or am I doing something wrong ??

[rif]

Without knowing more it's hard to say. The first thing to check is usually how your TFP code is running --- for instance if it's in eager mode it's going to be very slow. Can you share your code?

On Thu, Apr 7, 2022 at 10:49 PM Sarthak Kala <sarthak...@gmail.com> wrote:

Tried running a regression model with both TFP and PyStan. PyStan is atleast 10x faster. Is there a reason why? Or am I doing something wrong ??

--
You received this message because you are subscribed to the Google Groups "TensorFlow Probability" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tfprobabilit...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/tfprobability/a8d9bab2-e3ac-43fd-81d5-4e4fd363aaadn%40tensorflow.org.

[Sarthak Kala]

PyStan Code : https://colab.research.google.com/drive/1QwYPwBxy7hOYboW1_UScKhXB2aefUmnO?usp=sharing

TFP : https://colab.research.google.com/drive/1HQv0Zq_4c8Nw2bZfq_ICwujkur8W3lB1?usp=sharing

If anyone can help me in optimising my TFP code, that would be of great help. 

[Christiaan Swanepoel]

You are running your code in eager mode, which will be substantially slower. I ran your code in function mode and observed a ~30x speedup.

You can use function mode by wrapping your call to `tfp.mcmc.sample_chain` in a function without arguments and adding the `tf.function` decorator.