Local expert optimal interpolation

Bases: object

This provides the main interface for conducting an experiment in GPSat to predict an underlying field from satellite measurements using local Gaussian process (GP) models.

This proceeds by iterating over the local expert locations, training the local GPs on data in a neighbourhood of the expert location and making predictions on specified locations. The results will be saved in an HDF5 file.

Example usage:

>>> store_path = "/path/to/store.h5"
>>> locexp = LocalExpertOI(data_config, model_config, expert_loc_config, pred_loc_config)
>>> locexp.run(store_path=store_path) # Run full sweep and save results in store_path

Parameters:

expert_loc_config: dict or ExpertLocsConfig: Configuration for expert locations.
data_config: dict or DataConfig: Configuration for data to be interpolated.
model_config: dict or ModelConfig: Configuration for model used to perform the local optimal interpolation.
pred_loc_config: dict or PredictionLocsConfig: Configuration for prediction locations.
local_expert_config: ExperimentConfig, optional: If the above four configurations are stored in ExperimentConfig, you can pass this all at once by specifying a single ExperimentConfig.

Notes

See configuration dataclasses for more details on the specific configuration classes.

run(store_path=None, store_every=10, check_config_compatible=True, skip_valid_checks_on=None, optimise=True, predict=True, min_obs=3, table_suffix='')

Run a full sweep to perform local optimal interpolation at every expert location. The results will be stored in an HDF5 file containing (1) the predictions at each location, (2) parameters of the model at each location, (3) run details such as run times, and (4) the full experiment configuration.

Parameters:

store_path: str: File path where results should be stored as HDF5 file.
store_every: int, default 10: Results will be stored to file after every store_every expert locations. Reduce if optimisation is slow, must be greater than 1.
check_config_compatible: bool, default True: Check if current LocalExpertOI configuration is compatible with previous, if applicable. If file exists in store_path, it will check the oi_config attribute in the oi_config table to ensure that configurations are compatible.
skip_valid_checks_on: list, optional: When checking if config is compatible, skip keys specified in this list.
optimise: bool, default True: If True, will run model.optimise_parameters() to learn the model parameters at each expert location.
predict: bool, default True: If True, will run model.predict() to make predictions at the locations specified in the prediction locations configuration.
min_obs: int, default 3: Minimum number observations required to run optimisation or make predictions.
table_suffix: str, optional: Suffix to be appended to all table names when writing to file.

Returns:

None

Notes

By default, both training and inference are performed at every location. However one can opt to do either one with the optimise and predict options, respectively.
If check_config_compatible is set to True, it makes sure that all results saved to store_path use the same configurations. That is, if one re-runs an experiment with a different configuration but pointing to the same store_path, it will return an error. Make sure that if you run an experiment with a different configuration, either set a different store_path, or if you want to override the results, delete the generated store_path.
The table_suffix is useful for storing multiple results in a single HDF5 file, each with a different suffix. See <hyperparameter smoothing> for an example use case.

GPSat.local_experts.get_results_from_h5file(results_file, global_col_funcs=None, merge_on_expert_locations=True, select_tables=None, table_suffix='', add_suffix_to_table=True, verbose=False)

Retrieve results from an HDF5 file.

Parameters:

results_file: str: The location where the results file is saved. Must point to a HDF5 file with the file extension .h5.
select_tables: list, optional: A list of table names to select from the HDF5 file.
global_col_funcs: dict, optional: A dictionary of column functions to apply to selected tables.
merge_on_expert_locations: bool, default True: Whether to merge expert location data with results data.
table_suffix: str, optional: A suffix to add to selected table names.
add_suffix_to_table: bool, default True: Whether to add the table suffix to selected table names.
verbose: bool, default False: Set verbosity.

Returns:

tuple:

A tuple containing two elements:

dict: A dictionary of DataFrames where each table name is the key. This contains the predictions and learned model parameters at every location.
list: A list of configuration dictionaries.

Notes

This function reads data from an HDF5 file, applies optional column functions, and optionally merges expert location data with results data.
The 'select_tables' parameter allows you to choose specific tables from the HDF5 file.
Column functions specified in 'global_col_funcs' can be applied to selected tables.
Expert location data can be merged onto results data if 'merge_on_expert_locations' is set to True.