Skip to content
This repository has been archived by the owner on Sep 1, 2023. It is now read-only.

Delta encoder shouldn't use the Adaptive Scalar Encoder #3312

Open
BoltzmannBrain opened this issue Aug 30, 2016 · 0 comments
Open

Delta encoder shouldn't use the Adaptive Scalar Encoder #3312

BoltzmannBrain opened this issue Aug 30, 2016 · 0 comments

Comments

@BoltzmannBrain
Copy link
Contributor

The delta encoder should not subclass from and use the adaptive scalar encoder -- here. The adaptive scalar encoder can give poor results particularly for short data streams because the representations keep changing as it learns. It should be replaced with a fixed encoder or an RDSE.

While we're at it, the documentation in the delta encoder is lacking. For example, it's unclear that the minval and maxval params here should be the minimum and maximum deltas; e.g., if the scalar range is 2 to 82, the delta range can be from -80 to 80, so the range here should be twice that of a scalar encoder. This could be split into another issue, or included here as the constructor docstring is updated to reflect the new parent class.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Projects
None yet
Development

No branches or pull requests

2 participants