Is there any possibility of changing the dtype of a tensorflow_probability.distribution
, either on instantiation or with a respective setter()?
I want to use tf.float64
instead of the standard tf.float32
but could not find any way to do this, since the dtype
-attribute is not directly accessible and tf.cast()
does not work on it.
You generally just need to pass some numerical parameter explicitly, tf.cast or np.cast to f64.