Fitness Landscape Analysis for Neural Architecture Search:

Step 1: Install Nasbench-101 submodule

Generate descriptive data of NASbench-101:

• cd nasbench/
• python nasbench/nasalg/landscape.py > experiments/full-landscape.json

Analyze the Fitness Landscape of NASBench-101

• Statistics and FDC: nasbench/experiments/fitness-landscape.Rmd (R-sudio)
• Local optima estimation: card_optima/local_search_for_card_optima.py
• Local optima analysis: notebooks/Card_Optima_-CIFAR-_10.ipynb (notebook)
• Random Walk Analysis: notebooks/Random_Walks_Analysis_-_CIFAR-10.ipynb'(notebook)
• Persistence: notebooks/Persistence_-_CIFAR-10.ipynb'(notebook)

##Project using this code: Fitness Landscape Footprint: A Framework to Compare Neural Architecture Search Problems, arXiv:2111.01584, Kalifou René Traoré, Andrés Camero, Xiao Xiang Zhu.