CHANGELOG.md

3.0.0 (2021-11-17)

Refactor

• agent: explore/exploit methods does not return population

2.0.2 (2021-10-25)

Fix

• agent trial numbering starts from 1 (not 0)

2.0.1 (2021-10-25)

Fix

• yacs: prevention iterating empty expected improvements when specializing condition

2.0.0 (2021-10-25)

Fix

• epsilon value is drawn from uniform distribution
• epsilon value supports values from [0, 1]

Feat

• move the logic for selecting best action into each agent
• Use OpenAI Gym ObservationWrapper interface for transforming environment input

BREAKING CHANGE

• dropped support for environment_adapter configuration option

1.6 (2019-12-16)

1.5.0 (2019-08-07)

1.4.1 (2018-04-19)

1.4 (2018-04-19)

1.3 (2018-03-19)

1.2 (2017-09-08)

1.1 (2017-09-08)

1.0 (2017-09-08)