gt4sd.algorithms.conditional_generation.guacamol.implementation.smiles_lstm_ppo module

Recurrent Neural Networks with Proximal Policy Optimization algorithm implementation.

Summary

Classes:

SMILESLSTMPPO

Reference

class SMILESLSTMPPO(model_path, num_epochs, episode_size, optimize_batch_size, entropy_weight, kl_div_weight, clip_param)[source]

Bases: object

__init__(model_path, num_epochs, episode_size, optimize_batch_size, entropy_weight, kl_div_weight, clip_param)[source]

Initialize SMILESLSTMPPO.

Parameters

  • model_path (str) – path to load the model.

  • num_epochs (int) – number of epochs to sample.

  • episode_size (int) – number of molecules sampled by the policy at the start of a series of ppo updates.

  • optimize_batch_size (int) – batch size for the optimization.

  • entropy_weight (int) – used for calculating entropy loss.

  • kl_div_weight (int) – used for calculating Kullback-Leibler divergence loss.

  • clip_param (float) – used for determining how far the new policy is from the old one.

get_generator()[source]

Create an instance of the PPODirectedGenerator.

Return type

PPODirectedGenerator

Returns

an instance of PPODirectedGenerator.

__dict__ = mappingproxy({'__module__': 'gt4sd.algorithms.conditional_generation.guacamol.implementation.smiles_lstm_ppo', '__init__': <function SMILESLSTMPPO.__init__>, 'get_generator': <function SMILESLSTMPPO.get_generator>, '__dict__': <attribute '__dict__' of 'SMILESLSTMPPO' objects>, '__weakref__': <attribute '__weakref__' of 'SMILESLSTMPPO' objects>, '__doc__': None, '__annotations__': {}})
__doc__ = None
__module__ = 'gt4sd.algorithms.conditional_generation.guacamol.implementation.smiles_lstm_ppo'
__weakref__

list of weak references to the object (if defined)