Experiment Modules (Core)

Experiment Modules (Core)#

alts.core.experiment_modules
class EstimatorExperiment(query_selector, time_source, data_pools, poracles, estimator)[source]#

Bases: ExperimentModules

Description
EstimatorExperiment additionally has an estimator to train on the data.
Parameters:

  • query_selector (QuerySelector) – Decider over the most useful datapoints for continued learning

  • time_source (TimeSource) – A time source for the experiment

  • data_pools (DataPools) – Collector of already acquired data

  • oracles (Oracles) – The source of new data

  • estimator (Estimator) – An estimator that trains with the acquired data

estimator: Estimator = NOTSET#
post_init(self) → None[source]#
Description
Initializes the query_selector and estimator with its configuration.
class ExperimentModules(query_selector, time_source, data_pools, oracles)[source]#

Bases: Publisher

Description
ExperimentModules is a collection of configured modules necessary for an experiment.
Parameters:

  • query_selector (QuerySelector) – Decider over the most useful datapoints for continued learning

  • time_source (TimeSource) – A time source for the experiment

  • data_pools (DataPools) – Collector of already acquired data

  • oracles (Oracles) – The source of new data

data_pools: DataPools#
initialize(self) → None[source]#
Description
Initializes the experiment.
oracles: Oracles#
post_init(self) → None[source]#
Description
Initializes the query_selector with its configuration.
query_selector: QuerySelector = NOTSET#
step(self, iteration) → None[source]#
Description
Does a step in the experiment at time point iteration.
Parameters:

iteration (int) – Current iteration of the experiment

time_source: TimeSource#
class InitQueryEstimatorExperiment(query_selector, time_source, data_pools, poracles, initial_query_sampler, estimator)[source]#

Bases: InitQueryExperimentModules, EstimatorExperiment

Description
InitQueryEstimatorExperiment is an experiment setup with an initial query sample and an estimator to train.
Parameters:

  • query_selector (QuerySelector) – Decider over the most useful datapoints for continued learning

  • time_source (TimeSource) – A time source for the experiment

  • data_pools (DataPools) – Collector of already acquired data

  • oracles (POracles) – The source of new data

  • initial_query_sampler (QuerySampler) – A sampler to get the very first datapoints

  • estimator (Estimator) – An estimator that trains with the acquired data

class InitQueryExperimentModules(query_selector, time_source, data_pools, poracles, initial_query_sampler)[source]#

Bases: ExperimentModules

Description
InitQueryExperimentModules additionally has an initial QuerySampler and a Process Oracle.
Parameters:

  • query_selector (QuerySelector) – Decider over the most useful datapoints for continued learning

  • time_source (TimeSource) – A time source for the experiment

  • data_pools (DataPools) – Collector of already acquired data

  • oracles (POracles) – The source of new data

  • initial_query_sampler (QuerySampler) – A sampler to get the very first datapoints

init_queries(self) → None[source]#
Description
Samples the inital queries and adds them to the oracle.
Returns:

The sampled queries

Return type:

NDArray

initial_query_sampler: QuerySampler = NOTSET#
initialize(self) → None[source]#
Description
Initializes the experiment by sampling the first queries for the experiment.
oracles: POracles#
post_init(self) → None[source]#
Description
Initializes the query_selector and init_query_sampler with its configuration.
Raises:

TypeError – If oracles is not a POracle