Why does the RL Toolbox not support BatchNormalization layer?

Hi.

I'm currently trying DDPG with my own network.

But when I try to use BatchNormalizationLayer, the error message says Batch Normalization is not supported.

I don't see any theoretical reason the batch normalization cannot come in.

Is there any plan to update the toolbox to make it possible?

Or is there a specific reason that the batch normalization cannot come in?

Thanks.

[Abd-Elhamid Nabeel]

Actually i face the same problem, but i thought this would help :

1- calculate the running mean and running STD of the data in Simulink for example (or matlab)

2- make a matlab function in simulink/matlab that could calculate the zscore for each feature through the equation :

X_zs = (X - Xmean)/STD

This would standardize your data.

Answer 1

[Aditya Patil]

This is an known issue, and it might be fixed in any of the upcoming releases.

Answer 2

[Abd-Elhamid Nabeel]

Actually i face the same problem, but i thought this would help :

1- calculate the running mean and running STD of the data in Simulink for example (or matlab)

2- make a matlab function in simulink/matlab that could calculate the zscore for each feature through the equation :

X_zs = (X - Xmean)/STD

This would standardize your data.

Answer 3

[Maha Mosalam]

No one can reached to solution or update in new versions for that issue?