iicd/xautodl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix small bugs in DARTS-V1 for NASNet-Space

Browse Source
This commit is contained in:
D-X-Y 2020-01-17 22:28:33 +11:00
parent db2760c260
commit c2ff845d1b
3 changed files with 13 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
docs/ICLR-2019-DARTS.md
View File

@ -4,17 +4,22 @@ DARTS: Differentiable Architecture Search is accepted by ICLR 2019.
In this paper, Hanxiao proposed a differentiable neural architecture search method, named as DARTS.
Recently, DARTS becomes very popular due to its simplicity and performance.
**Run DARTS on the NAS-Bench-201 search space**:
## Run DARTS on the NAS-Bench-201 search space
```
CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/algos/DARTS-V2.sh cifar10 1 -1
CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/algos/GDAS.sh cifar10 1 -1
```
**Run the first-order DARTS on the NASNet search space**:
## Run the first-order DARTS on the NASNet/DARTS search space
This command will start to use the first-order DARTS to search architectures on the DARTS search space.
```
CUDA_VISIBLE_DEVICES=0 bash ./scripts-search/DARTS1V-search-NASNet-space.sh cifar10 -1
```
After searching, if you want to train the searched architecture found by the above scripts, you need to add the config of that architecture (will be printed in log) in [genotypes.py](https://github.com/D-X-Y/AutoDL-Projects/blob/master/lib/nas_infer_model/DXYs/genotypes.py).
In future, I will add a more eligent way to train the searched architecture from the DARTS search space.
# Citation
```

3
exps/algos/DARTS-V1.py
View File

@ -199,7 +199,8 @@ def main(xargs):
logger.log('<<<--->>> The {:}-th epoch : find the highest validation accuracy : {:.2f}%.'.format(epoch_str, valid_a_top1))
copy_checkpoint(model_base_path, model_best_path, logger)
with torch.no_grad():
logger.log('arch-parameters :\n{:}'.format( nn.functional.softmax(search_model.arch_parameters, dim=-1).cpu() ))
#logger.log('arch-parameters :\n{:}'.format( nn.functional.softmax(search_model.arch_parameters, dim=-1).cpu() ))
logger.log('{:}'.format(search_model.show_alphas()))
if api is not None: logger.log('{:}'.format(api.query_by_arch( genotypes[epoch] )))
# measure elapsed time
epoch_time.update(time.time() - start_time)

4
lib/models/cell_searchs/search_model_darts.py
View File

@ -53,6 +53,10 @@ class TinyNetworkDarts(nn.Module):
def get_alphas(self):
return [self.arch_parameters]
def show_alphas(self):
with torch.no_grad():
return 'arch-parameters :\n{:}'.format( nn.functional.softmax(self.arch_parameters, dim=-1).cpu() )
def get_message(self):
string = self.extra_repr()
for i, cell in enumerate(self.cells):