https://github.com/han-cai/PathLevel-EAS

Path-Level Network Transformation for Efficient Architecture Search, in ICML 2018.
https://github.com/han-cai/PathLevel-EAS

Last synced: about 1 month ago
JSON representation

Path-Level Network Transformation for Efficient Architecture Search, in ICML 2018.

• Host: GitHub
• URL: https://github.com/han-cai/PathLevel-EAS
• Owner: han-cai
• Created: 2018-06-03T03:25:54.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2018-06-12T03:15:44.000Z (over 6 years ago)
• Last Synced: 2024-08-01T08:06:41.714Z (4 months ago)
• Language: Python
• Homepage:
• Size: 61.5 KB
• Stars: 113
• Watchers: 5
• Forks: 21
• Open Issues: 5
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-automated-machine-learning - [code

README

        
# Path-Level Network Transformation for Efficient Architecture Search

Code for the paper [Path-Level Network Transformation for Efficient Architecture Search](https://arxiv.org/abs/1806.02639) in ICML 2018. 

# Reference

```bash

@article{cai2018path,

  title={Path-Level Network Transformation for Efficient Architecture Search},

  author={Cai, Han and Yang, Jiacheng and Zhang, Weinan and Han, Song and Yu, Yong},

  journal={arXiv preprint arXiv:1806.02639},

  year={2018}

}

```

## Related Projects

- [Efficient Architecture Search by Network Transformation](https://arxiv.org/abs/1707.04873) in AAAI 2018, [Code](https://github.com/han-cai/EAS).

## Dependencies

* Python 3.6 

* Pytorch 0.3.1

## Results

### CIFAR-10

|           Model          | Params | Test error (%) | 

| ----------------------- | ------------- | ----- |

| [TreeCell-A with DenseNet (N=16, k=48, G=2)](https://drive.google.com/file/d/19OzTjozcJlbP4SZJXPiuQfZgklcs7nCX/view?usp=sharing) |  13.1M | 3.35 |

| [TreeCell-A with PyramidNet (N=18, alpha=84, G=2)](https://drive.google.com/open?id=1TgBI5y_j3YjTemqCalOPn3dA7CChZ1bn) | 5.7M | 3.14 |

| [TreeCell-A with PyramidNet (N=18, alpha=84, G=2) + DropPath (600 epochs)](https://drive.google.com/open?id=1AIwRxNrpX9N2GcsB6PXVfKgYaQnxKDF5) | 5.7M | 2.99 |

| [TreeCell-A with PyramidNet (N=18, alpha=84, G=2) + DropPath + Cutout (600 epochs)](https://drive.google.com/open?id=1BFcB9iaCWX8QgmhbGwqc3ZRZ9d61CJpo) | 5.7M | 2.49 |

| [TreeCell-A with PyramidNet (N=18, alpha=150, G=2) + DropPath + Cutout (600 epochs)](https://drive.google.com/open?id=1WbI4fE-m7f2leLR8nJMC1Cby4fWHFRUa) | 14.3M | 2.30 |

For checking these networks, please download the corresponding model files and run the following command under the folder of **code/CIFAR**:

```bash

$ python3 run_exp.py --path 

```

For example, by running

```bash

$ python3 run_exp.py --path ../../Nets/CIFAR10#PyramidTreeCellA#N=18_alpha=150#600#cutout

```

you will get

```bash

test_loss: 0.092100	 test_acc: 97.700000

```

### ImageNet

|           Model          | Multi-Add | Top-1 error (%) | 

| ----------------------- | ------------- | ----- |

| [TreeCell-B with CondenseNet (G1=4, G3=8)](https://drive.google.com/open?id=1BD3VdUzStaXiipA8oXM3SPWBUnuPtfjk)| 594M | 25.4 | 

Please refer to the file **code/ImageNet/scripts.sh**.

## Architecture Search

For setting up your environment to run architecture search experiments, please refer to my previous [repository](https://github.com/han-cai/EAS/tree/master/code).