LCMNAS

Instalation:

sudo apt-get install python3-dev graphviz libgraphviz-dev pkg-config python3 -m venv venv source venv/bin/activate pip install wheel torchcontrib pytorchcv pip install -r requirements.txt

Get Data sets (informations are anonymized)

Download data.zip and extract it in the main directory of the code from https://drive.google.com/drive/folders/1bnXQfYfNyYKCSX05DSFG8gE---EkfXSP?usp=sharing

Get Best Architectures Found (informations are anonymized)

Download models.zip (these do not have the autoaugment training) from https://drive.google.com/drive/folders/1bnXQfYfNyYKCSX05DSFG8gE---EkfXSP?usp=sharing

Experiments (the results get automatically stored in the experiments folder)

Mixed-performance estimation mechanism: (lambda=0.75)

python3 count.py --generations 50 --population 100 --dataset cifar10 --batch_size 96 —-mixed_training --mixed_fitness_lambda 0.75

Using only zero-proxy estimation (lambda=1)

python3 count.py --generations 50 --population 100 --dataset cifar10 --batch_size 96 --without_training --mixed_fitness_lambda 1

Using only regular training (lambda=0)

python3 count.py --generations 50 --population 100 --dataset cifar10 --batch_size 96 --mixed_fitness_lambda 0

Available datasets: cifar10 cifar100 ImageNet16-120

python3 count.py --generations 50 --population 100 --dataset cifar10 --batch_size 96 —-mixed_training --mixed_fitness_lambda 0.75

python3 count.py --generations 50 --population 100 --dataset cifar100 --batch_size 96 —-mixed_training --mixed_fitness_lambda 0.75

python3 count.py --generations 50 --population 100 --dataset ImageNet16 --batch_size 96 —-mixed_training --mixed_fitness_lambda 0.75

Train with auto-augment

python aux.py --searched_dataset cifar10 --dataset cifar10 --auto_augment --epochs 1500 --reset_weight --model_path "path_to_model"

Transfer to other data set

python aux.py --searched_dataset cifar10 --dataset ImageNet16 --reset_weight --model_path "path_to_model"

Thank you for the time spent reviewing our paper.