We are filtering out matrices with <7 nodes.

Multi-Objective-NAS / self-supervised-nas

Official implementation of the paper "Pretraining Neural Architecture Search Controllers with Locality-based Self-Supervised Learning" (NeurIPSW 2020)

5 stars 0 forks source link

We are filtering out matrices with <7 nodes. #15

Closed juice500ml closed 3 years ago

juice500ml commented 3 years ago

As-is.

https://github.com/Multi-Objective-NAS/self-supervised-nas/blob/eede911/src/datasets.py#L64
Currently, we are only covering size == 7 matrices, which covers about 70%.
- I'm not sure about the exact ratio. @Naruu knows about this.

To-be.

It is somewhat irrational to filter out <7. We are going to cover all the matrices.
@young917 is trying to solve this with multi-length sequence.

Related. issue #8

Ref. torch.nn.utils.rnn.pack_sequence

juice500ml commented 3 years ago

Solved in ab81d875c053ce16f3c99d1c57fcb323fc292cf9.