Big fan of the library. I was hoping to create a model like this one from here and was wondering if you plan on making any features analogous to those in tensorflow_probability.layers?
model = tfk.Sequential([
tfkl.Dense(tfpl.MultivariateNormalTriL.params_size(d)),
tfpl.MultivariateNormalTriL(d),
])
I have a haiku / distrax implementation of the above layer, but I am missing something to consistently constrain "allowable" the covariance matrix (full) and mean of the Gaussian distribution.
Hi,
Big fan of the library. I was hoping to create a model like this one from here and was wondering if you plan on making any features analogous to those in
tensorflow_probability.layers
?I have a
haiku
/distrax
implementation of the above layer, but I am missing something to consistently constrain "allowable" the covariance matrix (full) and mean of the Gaussian distribution.Cheers