search

D-X-Y / AutoDL-Projects

Automated deep learning algorithms implemented in PyTorch.
MIT License
1.56k stars 281 forks source link

How to train the model I searched? #40

Closed Debrove closed 4 years ago

Debrove commented 4 years ago

Thanks for your excellent works.

In SETN, as a newcomer to NAS, I am confused that I can't find a way to train the model I have searched. I run the scripts of training and searching respectively, which just train the default searched model instead of my searched model.

Thank you.

D-X-Y commented 4 years ago

Do you mean you want to train the model searched by these scripts (https://github.com/D-X-Y/NAS-Projects/tree/master/scripts-search/algos)?

Debrove commented 4 years ago

Yes.

shashank3959 commented 4 years ago

I'm currently deep into the code too trying to train the model I searched. The file https://github.com/D-X-Y/NAS-Projects/blob/master/lib/nas_infer_model/DXYs/genotypes.py looks like it has manually populated the searched architecture. If we have searched our own architecture, shall we make a similar entry manually?

shashank3959 commented 4 years ago

This is really good work btw!

Another question I have is that in GDAS, do we assume a fixed reduction cell as opposed to searching for one?

D-X-Y commented 4 years ago

@Debrove Thanks for your interest. This repo contains many algorithms. I will prepare script to train the searched model soon.

@shashank3959 In GDAS, we can search for both normal/reduction cells or only search for the nomral-cell with a fixed reduction cell. If you follow the instruction in README, it will automatically search for a tiny network instead of the structure in the paper. This is for fairly compare ours with other 10 NAS algorithms.

D-X-Y commented 4 years ago

@Debrove Thanks a lot for your interest. I will update the codes right after Dec 20. Due to some restrictions, I cannot release parts of the code at the moment. Sorry for the inconvenience.

Debrove commented 4 years ago

@D-X-Y Thank you very much.

D-X-Y commented 4 years ago

@Debrove I have updated the README. You can follow https://github.com/D-X-Y/NAS-Projects#training-the-searched-architecture to train your searched model. BTW: the current search space in this repo is the one in our recent paper "NAS-Bench-102: Extending the Scope of Reproducible Neural Architecture Search".