Closed skeydan closed 4 years ago
as per discussions on the TFP mailing list:
We recommend using tfb.AutoregressiveNetwork over tfb.masked_autoregressive_default_template. AutoregressiveNetwork is an updated version designed to work with TF 2.0 and Keras. There is an example of training a MaskedAutoregressiveFlow bijector with an AutoregressiveNetwork (using the Keras Functional API) in the docstring for the AutoregressiveNetwork class at https://github.com/tensorflow/probability/blob/56a11e4152dd5ac7d67e8cdefbbfb2001c89e11a/tensorflow_probability/python/bijectors/masked_autoregressive.py#L593 .
We recommend using tfb.AutoregressiveNetwork over tfb.masked_autoregressive_default_template. AutoregressiveNetwork is an updated version designed to work with TF 2.0 and Keras.
tfb.AutoregressiveNetwork
tfb.masked_autoregressive_default_template
There is an example of training a MaskedAutoregressiveFlow bijector with an AutoregressiveNetwork (using the Keras Functional API) in the docstring for the AutoregressiveNetwork class at https://github.com/tensorflow/probability/blob/56a11e4152dd5ac7d67e8cdefbbfb2001c89e11a/tensorflow_probability/python/bijectors/masked_autoregressive.py#L593 .
as per discussions on the TFP mailing list: