pytorch_simple_DARTS
pytorch_simple_DARTS copied to clipboard
A simple pytorch implementation of Differentiable Architecture Search (DARTS)
A simple Pytorch implementation of Differentiable Architecture Search (DARTS)
This repository is my pytorch implementation of Differentiable Architecture Search (DARTS). Some of the code is taken from the offical implementation.
Requirements:
- python >= 3.5
- pytorch >= 1.0
- tensorboardX (optional)
Search
python3 cifar_search.py --log_name darts_cifar_search --order 1st --gpus 0
Evaluate
python cifar_eval.py --log_name darts_cifar_search --gpus 0python imgnet_eval.py --log_name darts_cifar_search --data_dir YOUR_IMGNET_DIR --gpus 0,1,2,3
CIFAR10 Results:
| Method | Acc. |
|---|---|
| DARTS 1st order | 97.06% |
| DARTS 2st order | 97.36% |