google-deepmind / google-deepmind/distrax
MADE masking like tfb.AutoregressiveNetwork
Open
- Dominant language
- Python
- Stars
- 651
- Forks
- 48
- Avg merge
- 21h 4m
- Merged PRs (30d)
- 3
Description
Hello,
Am I right to say that SplitCoupling seems related to the Real NVP simple coupling, and MaskCoupling can implement a chessboard coupling?
And what about the autoregressive MADE like masking strategy, is it implemented somewhere?
I am looking for an equivalent of TF bijectors
```python
hidden_units = [128, 128]
bijectors.append(tfb.MaskedAutoregressiveFlow( name ='MAF%d' %i,
shift_and_log_scale_fn=tfb.AutoregressiveNetwork(
params=2, hidden_units=hidden_units, activation=tf.nn.leaky_relu)))
```
Thanks
Contributor guide
Assessment
This issue has not been assessed yet.