API¶

Assistant:¶

class labwatch.assistant.LabAssistant(experiment, database_name=None, url=’localhost’, optimizer=None, prefix=’runs’, always_inject_observer=False)¶

The main class for Labwatch. It runs an experiment with a configuration suggested by and hyperparameter optimizer.

The hyperparameter optimizer uses the information about the experiment that are stored in the database to suggest a new configuration.

search_space(function)¶

Decorator for creating a searchspace definition from a function.

Hyperparameters:¶

class labwatch.hyperparameters.Gaussian(mu, sigma, log_scale=False, uid=None)¶

A Gaussian just has a different distribution

labwatch.hyperparameters.decode_param_or_op(storage)¶

Decode method for converting BSON like dicts to parameter values.

class labwatch.hyperparameters.Parameter(uid=None, fixed=None)¶

class labwatch.hyperparameters.Constant(value, uid=None)¶

class labwatch.hyperparameters.Categorical(choices_in, uid=None)¶

class labwatch.hyperparameters.UniformNumber(lower, upper, type, default=None, log_scale=False, uid=None)¶

class labwatch.hyperparameters.UniformFloat(lower, upper, default=None, log_scale=False, uid=None)¶

class labwatch.hyperparameters.UniformInt(lower, upper, default=None, log_scale=False, uid=None)¶

class labwatch.hyperparameters.Gaussian(mu, sigma, log_scale=False, uid=None)

A Gaussian just has a different distribution

class labwatch.hyperparameters.ConditionResult(result, condition)¶

class labwatch.hyperparameters.Condition(param, choices)¶

Converters:¶

labwatch.converters.convert_to_configspace.configspace_config_to_sacred(config)¶

Convert a Configuration into a dict mapping parameter names to values.

config: ConfigSpace.Configuration

dict:

Dictionary mapping parameter names to values.

labwatch.converters.convert_to_configspace.convert_simple_param(name, param)¶

Convert a simple labwatch parameter to a ConfigSpace parameter.

name: str

The name of the parameter.

param: dict

Dictionary describing the parameter.

ConfigSpace.hyperparameters.Hyperparameter:

The converted hyperparameter.

labwatch.converters.convert_to_configspace.sacred_config_to_configspace(cspace, config)¶

Fill a ConfigurationSpace with the given values and return the resulting Configuration.

cspace: ConfigSpace.ConfigurationSpace

The configuration space to be populated.

config: dict

The configuration values as a dictionary mapping names to values.

ConfigSpace.Configuration:

The resulting Configuration.

labwatch.converters.convert_to_configspace.sacred_space_to_configspace(space)¶

Convert a Labwatch searchspace to a ConfigSpace.

space: labwatch.searchspace.SearchSpace

A labwatch searchspace to be converted.

ConfigSpace.ConfigurationSpace:

A ConfigurationSpace equivalent to the given SeachSpace.

Optimizers:¶

class labwatch.optimizers.base.Optimizer(config_space)¶

Defines the interface for all optimizers.

needs_updates()¶

bool:

True if this optimizer needs updates, False otherwise.

suggest_configuration()¶

Suggests a configuration of hyperparameters to be run.

dict:

Dictionary mapping parameter names to suggested values.

The default is to return nothing this is done such that the user / the watcher can then check the queue or try something else. If you want get_random_config as default create an instance of RandomSearch instead.

update(configs, costs, runs)¶

Update the internal state of the optimizer with a list of new results.

configs: list[dict]

List of configurations mapping parameter names to values.

costs: list[float]

List of costs associated to each config.

runs: list[dict]

List of dictionaries containing additional run information.

class labwatch.optimizers.bayesian_optimization.BayesianOptimization(config_space, burnin=100, chain_length=200, n_hypers=20)¶

class labwatch.optimizers.bohamiann.Bohamiann(config_space, burnin=3000, n_iters=10000)¶

class labwatch.optimizers.dngo.DNGOWrapper(config_space, burnin=1000, chain_length=200, n_hypers=20)¶

class labwatch.optimizers.random_search.RandomSearch(config_space)¶