runnn

run neural networks

yet another runner for neural network training

Core Ideas:

• support multiple workers
• params are stored together with final test score
• support parameter search (sampling, grid search)
• analyze data with sql (no log files)

Usage

run a script 10 times python3 server.py run _command_ --tag=test_run --num_iters=10

start a worker python3 client.py

check scores python3 server.py get_scores --tag=test_run

analyze results python3 server.py sql --cmd 'SELECT log FROM tasks WHERE id_task = 47'

python3 server.py sql --cmd 'SELECT log FROM tasks WHERE score > 0.8'

python3 server.py sql --cmd 'SELECT score FROM tasks WHERE id_run = 15 and cmd LIKE "%learning_rate=0.001%"' Results will be printed to the console. Any sql query is possible!

credits

inspired by https://github.com/haeusser/NebuLight and https://github.com/haeusser/runjob