iicd/xautodl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
5c73aeb50b
xautodl/AA-NAS-Bench.md
D-X-Y 7b354d4c74 updates
2019-11-11 00:46:02 +11:00

3.5 KiB
Raw Blame History

An Algorithm-Agnostic NAS Benchmark (AA-NAS-Bench)

We propose an Algorithm-Agnostic NAS Benchmark (AA-NAS-Bench) with a fixed search space, which provides a unified benchmark for almost any up-to-date NAS algorithms. The design of our search space is inspired from that used in the most popular cell-based searching algorithms, where a cell is represented as a directed acyclic graph. Each edge here is associated with an operation selected from a predefined operation set. For it to be applicable for all NAS algorithms, the search space defined in AA-NAS-Bench includes 4 nodes and 5 associated operation options, which generates 15,625 neural cell candidates in total.

In this Markdown file, we provide:

  • Detailed instruction to reproduce AA-NAS-Bench.
  • 10 NAS algorithms evaluated in our paper.

Note: please use PyTorch >= 1.1.0 and Python >= 3.6.0.

Instruction to Generate AA-NAS-Bench

  1. generate the meta file for AA-NAS-Bench using the following script, where AA-NAS-BENCH indicates the name and 4 indicates the maximum number of nodes in a cell.
bash scripts-search/AA-NAS-meta-gen.sh AA-NAS-BENCH 4
  1. train earch architecture on a single GPU (see commands in output/AA-NAS-BENCH-4/meta-node-4.opt-script.txt which is automatically generated by step-1).
CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/AA-NAS-train-archs.sh     0   389 -1 '777 888 999'

This command will train 390 architectures (id from 0 to 389) using the following four kinds of splits with three random seeds (777, 888, 999).

Dataset Train Eval
CIFAR-10 train valid
CIFAR-10 train + valid test
CIFAR-100 train valid+test
ImageNet-16-120 train valid+test
  1. calculate the latency, merge the results of all architectures, and simplify the results. (see commands in output/AA-NAS-BENCH-4/meta-node-4.cal-script.txt which is automatically generated by step-1).
OMP_NUM_THREADS=6 CUDA_VISIBLE_DEVICES=0 python exps/AA-NAS-statistics.py --mode cal --target_dir 000000-000389-C16-N5
  1. merge all results into a single file for AA-NAS-Bench-API.
OMP_NUM_THREADS=4 python exps/AA-NAS-statistics.py --mode merge

[option] train a single architecture on a single GPU.

CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/AA-NAS-train-net.sh resnet 16 5
CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/AA-NAS-train-net.sh '|nor_conv_3x3~0|+|nor_conv_3x3~0|nor_conv_3x3~1|+|skip_connect~0|skip_connect~1|skip_connect~2|' 16 5

[option] load the parameters of a trained network.

To reproduce 10 baseline NAS algorithms in AA-NAS-Bench

We have tried our best to implement each method. However, still, some algorithms might obtain non-optimal results since their hyper-parameters might not fit our AA-NAS-Bench. If researchers can provide better results with different hyper-parameters, we are happy to update results according to the new experimental results. We also welcome more NAS algorithms to test on our dataset and would include them accordingly.

  • CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/algos/DARTS-V1.sh cifar10 -1
  • CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/algos/DARTS-V2.sh cifar10 -1
  • CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/algos/GDAS.sh cifar10 -1
  • CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/algos/SETN.sh cifar10 -1
  • CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/algos/ENAS.sh cifar10 -1
  • CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/algos/RANDOM-NAS.sh cifar10 -1