Open izsahara opened 3 years ago
The error is triggered when save_weights_only=False
. For some reason this ends up calling the likelihood layer with a Tensor rather than a MultivariateNormalDiag...
Thanks for raising the issue. This is a known bug: it is currently not possible to persist GPflux models, which rely on tfp.layers.DistributionLambda
layers, beyond just storing the weights with save_weights_only=True
. I will leave this issue open for better visibility.
Is there any solution yet ?
Describe the bug Literally the title
The Error/Console Output
System information