The following code runs fine for a standard neural network, but not for a Bayesian neural network. In the latter case, calling the function fitleads R to abort, not before saying that tensorflow/compiler/tf2mlcompute/kernels/mlc_subgraph_op.cc Compute: Failed in processing TF graph gradient tape.
I would appreciate if you could share any pointers into how to make this work. Could this be an issue related to Apple's Tensorflow implementation? Or just some wrong code from my side? Finally, could this be a bug in the R implementation of tfprobability?
Hi all,
The following code runs fine for a standard neural network, but not for a Bayesian neural network. In the latter case, calling the function
fit
leads R to abort, not before saying that tensorflow/compiler/tf2mlcompute/kernels/mlc_subgraph_op.cc Compute: Failed in processing TF graph gradient tape.I would appreciate if you could share any pointers into how to make this work. Could this be an issue related to Apple's Tensorflow implementation? Or just some wrong code from my side? Finally, could this be a bug in the R implementation of tfprobability?
Best regards and thanks in advance.