Skip to content

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

Notifications You must be signed in to change notification settings

Multi-Objective-NAS/self-supervised-nas

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

76 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Pretraining Neural Architecture Search Controllers with Locality-based Self-Supervised Learning

Environments

  • conda environment configs: environment.yaml
  • How to install:
conda env create -f environment.yaml
conda activate self-supervised-nas-3.6
pip install -e libs/nasbench --no-deps
pip install -e libs/NAS-Bench-201

How to train

  • CUDA_VISIBLE_DEVICES={device_index} python3 pretrain.py experiment=TripletMarginLoss
  • outputs/{date}/{time}/logs -> tensorboard directory
  • Above directory can be checked with tensorboard --bind_all --logdir=/path/to/log --port 8080 or tensorboard --bind_all --logdir_spec=exp1:/path/to/log1,exp2:/path/to/log2 --port 8080
  • outputs/{date}/{time}/weights -> output model weights
  • Use with caution pgrep -f "python pretrain.py" -a will show all the training process. pkill -f "python pretrain.py" will kill them all.

tmux cheatsheet

  • tmux ls: Current tmux sessions
  • tmux new -s {session_name}: Create session with session_name
  • tmux attach -t {session_name}: Attach to existing session
  • (Inside tmux session) Ctrl+a d: Detach existing session
  • (Inside tmux session) Ctrl+a PgUp/PgDn: Scroll inside session (Ctrl+c to abort)
  • (Inside tmux session) Ctrl+d: Close session

About

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

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages