Pre-built Environments

This module contains several pre-built experiments. Each pre-built environment is created by overriding the following components:

• Action Handler refers to overriding of the _step() method of ConformerEnv, which determines how the molecule is modified given some action.

• Reward Handler refers to overriding of the _reward() method of ConformerEnv, which determines how the reward is calculated based on the current configuration of the molecule.

• Observation Handler refers to overriding of the _obs() method of ConformerEnv, which returns an observation object based on the current configuration of the molecule and is a compatible input for the neural net being used for training.

All pre-built environments inherit from ConformerEnv and share the same constructor.

class conformer_rl.environments.environments.DiscreteActionEnv(mol_config: conformer_rl.config.mol_config.MolConfig)

• Action Handler: DiscreteActionMixin

• Reward Handler: default reward handler from ConformerEnv

• Observation Handler: GraphObsMixin

class conformer_rl.environments.environments.GibbsScoreEnv(mol_config: conformer_rl.config.mol_config.MolConfig)

• Action Handler: DiscreteActionMixin

• Reward Handler: GibbsRewardMixin

• Observation Handler: AtomCoordsTypeGraphObsMixin

class conformer_rl.environments.environments.GibbsScorePruningEnv(mol_config: conformer_rl.config.mol_config.MolConfig)

• Action Handler: DiscreteActionMixin

• Reward Handler: GibbsPruningRewardMixin

• Observation Handler: AtomCoordsTypeGraphObsMixin

class conformer_rl.environments.environments.GibbsScoreEndPruningEnv(mol_config: conformer_rl.config.mol_config.MolConfig)

• Action Handler: DiscreteActionMixin

• Reward Handler: GibbsEndPruningRewardMixin

• Observation Handler: AtomCoordsTypeGraphObsMixin

class conformer_rl.environments.environments.GibbsScoreLogPruningEnv(mol_config: conformer_rl.config.mol_config.MolConfig)

• Action Handler: DiscreteActionMixin

• Reward Handler: GibbsLogPruningRewardMixin

• Observation Handler: AtomCoordsTypeGraphObsMixin

class conformer_rl.environments.environments.GibbsScorePruningCurriculumEnv(mol_configs: List[conformer_rl.config.mol_config.MolConfig])

Same handlers as the GibbsScorePruningEnv but with support for curriculum learning.

class conformer_rl.environments.environments.GibbsScoreLogPruningCurriculumEnv(mol_configs: List[conformer_rl.config.mol_config.MolConfig])

Same handlers as the GibbsScoreLogPruningEnv but with support for curriculum learning.