Are there any plans to depend on tensorflow?

I recently read a HacknerNews thread on Stan and probabilistic programming more generally and one user commented that

“The next cut of Stan will likely use TFP [Tensorflow Probability] as a backend”

I see that the thread is from February 2021 so I was wondering firstly whether this was a plan or an opinion and secondly whether the topic has moved on since then?

There is no plan for Stan to depend on tensorflow, though there is an experimental version of the stanc3 compiler which can compile to tensorflow probability



1 Like

It’s slower than using the stan math library and was made because tfp was interested in Stan as a frontend


the TFP backend was discussed at the most recent Stan Language dev meeting: Monthly language dev meeting - #22 by Bob_Carpenter