sergia-ch/gin_tune

Integration between gin-config and ray tune

Integration between tune and gin-config

It is awesome to select hyperparameters automatically with tune. It is cool to specify configuration using gin-config.

It is even better to specify parameters for your tune runs via gin!

Install:

pip install -e .

Usage example

See small_test directory

config.gin

import myfunc
import gin_tune

tune1/grid_search.values = [123, 456]
tune2/grid_search.values = [789, 12]

myfunc.f.x1 = @tune1/grid_search()
myfunc.f.x2 = @tune2/grid_search()

gin_tune_config.num_workers = 0
tune_run.verbose = True

myfunc.py

import gin
from ray import tune

@gin.configurable
def f(config, x1, x2):
    """Example function to tune."""
    tune.report(sum=x1+x2)

tune.py

from gin_tune import tune_gin
from myfunc import f
import gin


if __name__ == '__main__':
    gin.parse_config_file('config.gin')
    analysis = tune_gin(f)

    print("Sum results")
    print([(x, y['sum']) for x, y in analysis.results.items()])

Run python tune.py. Should print [135, 468, 912, 1245].

Files test_config.py and test.gin in the gin_tune_tests folder contain a more complicated example.

Tests

Just run pytest