wallarm / nascell-automl

nascell-automl

This code belongs to the "Simple implementation of Neural Architecture Search with Reinforcement Learning " blog post.

An original blog post with all the details (step-by-step guide): https://lab.wallarm.com/the-first-step-by-step-guide-for-implementing-neural-architecture-search-with-reinforcement-99ade71b3d28

Requirements

• Python 3
• Tensorflow > 1.4

Training

Print parameters:

python3 train.py --help

optional arguments:
  -h, --help            show this help message and exit
  --max_layers MAX_LAYERS

Train:

python3 train.py

For evaluate architecture

Print parameters:

$ cd experiments/
$ python3 train.py --architecture "61, 24, 60,  5, 57, 55, 59, 3"