gt4sd.algorithms.controlled_sampling.paccmann_gp.implementation module¶
Implementation of PaccMann^GP conditional generator.
Summary¶
Classes:
Conditional generator as implemented in https://doi.org/10.1021/acs.jcim.1c00889. |
Reference¶
- class GPConditionalGenerator(resources_path, temperature=1.4, generated_length=100, batch_size=32, limit=5.0, acquisition_function='EI', number_of_steps=32, number_of_initial_points=16, initial_point_generator='random', seed=42, number_of_optimization_rounds=1, sampling_variance=0.1, samples_for_evaluation=4, maximum_number_of_sampling_steps=32, device=None)[source]¶
Bases:
objectConditional generator as implemented in https://doi.org/10.1021/acs.jcim.1c00889.
- __init__(resources_path, temperature=1.4, generated_length=100, batch_size=32, limit=5.0, acquisition_function='EI', number_of_steps=32, number_of_initial_points=16, initial_point_generator='random', seed=42, number_of_optimization_rounds=1, sampling_variance=0.1, samples_for_evaluation=4, maximum_number_of_sampling_steps=32, device=None)[source]¶
Initialize the conditional generator.
- Parameters
resources_path (
str) – directory where to find models and parameters.temperature (
float) – temperature parameter for the softmax sampling in decoding. Defaults to 1.4.generated_length (
int) – maximum length in tokens of the generated molcules (relates to the SMILES length). Defaults to 100.batch_size (
int) – batch size used for the generative model sampling. Defaults to 16.limit (
float) – hypercube limits in the latent space. Defaults to 5.0.acquisition_function (
str) – acquisition function used in the Gaussian process. Defaults to “EI”. More details in https://scikit-optimize.github.io/stable/modules/generated/skopt.gp_minimize.html.number_of_steps (
int) – number of steps for an optmization round. Defaults to 32.number_of_initial_points (
int) – number of initial points evaluated. Defaults to 16.initial_point_generator (
str) – scheme to generate initial points. Defaults to “random”. More details in https://scikit-optimize.github.io/stable/modules/generated/skopt.gp_minimize.html.seed (
int) – seed used for random number generation in the optimizer. Defaults to 42.number_of_optimization_rounds (
int) – maximum number of optimization rounds. Defaults to 1.sampling_variance (
float) – variance of the Gaussian noise applied during sampling from the optimal point. Defaults to 0.1.samples_for_evaluation (
int) – number of samples averaged for each minimization function evaluation. Defaults to 4.maximum_number_of_sampling_steps (
int) – maximum number of sampling steps in an optmization round. Defaults to 32.device (
Union[device,str,None]) – . Defaults to None, a.k.a, picking a default one (“gpu” if present, “cpu” otherwise).
- target_to_minimization_function(target)[source]¶
Use the target to configure a minimization function.
- Parameters
target (
Union[Dict[str,Dict[str,Any]],str]) – dictionary or JSON string describing the optimization target.- Return type
CombinedMinimization- Returns
a minimization function.
- __dict__ = mappingproxy({'__module__': 'gt4sd.algorithms.controlled_sampling.paccmann_gp.implementation', '__doc__': 'Conditional generator as implemented in https://doi.org/10.1021/acs.jcim.1c00889.', '__init__': <function GPConditionalGenerator.__init__>, 'target_to_minimization_function': <function GPConditionalGenerator.target_to_minimization_function>, 'set_seed': <function GPConditionalGenerator.set_seed>, 'generate_batch': <function GPConditionalGenerator.generate_batch>, '__dict__': <attribute '__dict__' of 'GPConditionalGenerator' objects>, '__weakref__': <attribute '__weakref__' of 'GPConditionalGenerator' objects>, '__annotations__': {}})¶
- __doc__ = 'Conditional generator as implemented in https://doi.org/10.1021/acs.jcim.1c00889.'¶
- __module__ = 'gt4sd.algorithms.controlled_sampling.paccmann_gp.implementation'¶
- __weakref__¶
list of weak references to the object (if defined)