Eugen Beck

There is one remaining warning that is not easily fixable: I import _SubnetworkRecCell to do an isinstance check. Can this warning be ignored or how should I fix that?

We use RETURNN also on some Centos 7 / Amazon EC2 Linux machines with python 3.6. It would be nice if support for python 3.6 remained let's say until the...

Actually the one EC2-Linux machine I checked had python3.7 but that one was ARM based. I'm more concerned about the whole Centos situation. Redhat stopped support for Centos stable releases...

There is a version of RASR that uses python3 instead of python2 and I can port it to the public version if there is demand for that.

We (at i6) have not yet tried to convert our models to TFLite. My suggestion would be to freeze the graph and then use the tools that come with TFLite...

Regarding Lingvo: yes, the project is older than the tf.data API

Commenting on your 3 suggestions in order: 1. Not very scalable, having to add this to every dataset makes it unnecessarily complicated. 2. The dataset classes contain more logic than...

Regarding 1: OK, I guess we agree not to do it this way. Regarding 2: My main worry is that readability will suffer if we have to many nested datasets,...

We definetly need to distinguish train and dev datasets. If we do data augmentation for training we don't necessarily want to do it for cross-validation. If we are doing some...

Could we add this to (every) engine class? The engine knows what kind of task it performs and what dataloader it uses for that task and could pick the correct...