Optimization

Grid search optimization over DR hyperparameters.

Parameter Classes

cdr_bench.optimization.params.DimReducerParams dataclass

cdr_bench.optimization.params.PCAParams dataclass

Bases: DimReducerParams

cdr_bench.optimization.params.UMAPParams dataclass

Bases: DimReducerParams

cdr_bench.optimization.params.TSNEParams dataclass

Bases: DimReducerParams

cdr_bench.optimization.params.GTMParams dataclass

Bases: DimReducerParams

Class for keeping track of GTM-specific parameters, inheriting from DimReducerParams.

default_params(n_components) staticmethod

Returns default parameters for GTM based on the number of components.

Parameters:

Name	Type	Description	Default
n_components	int	Number of components for GTM.	required

Returns:

Type	Description
dict[str, Any]	Dict[str, Any]: A dictionary of default parameters.

Raises:

Type	Description
ValueError	If n_components is not 2 or 3.

cdr_bench.optimization.params.ScoringParams dataclass

A data class to store parameters for scoring.

Attributes:

Name	Type	Description
ambient_dim_indices	List	Indices of the high-dimensional data.
n_neighbors	int	Number of neighbors to consider.
other_param	Any	Other optional parameters.
split	Any	Split parameter for compatibility with scikit-learn.
low_dim_metric	str	Metric for the low-dimensional space.
normalize	Any	Flag to normalize the score.

from_neighbors_indices(k_neighbors, indices) classmethod

Create ScoringParams from nearest neighbors indices.

Parameters:

Name	Type	Description	Default
k_neighbors	int	Number of nearest neighbors to consider.	required
indices	ndarray	Indices of the nearest neighbors.	required

Returns:

Name	Type	Description
ScoringParams	ScoringParams	An instance of ScoringParams with the calculated parameters.

cdr_bench.optimization.params.OptimizerParams dataclass

Optimizer

cdr_bench.optimization.optimization.Optimizer

__init__(params)

grid_search(X, y=None)

Perform grid search over parameter grid, fit estimator to data, and evaluate performance.

Parameters:

Name	Type	Description	Default
X	ndarray	Training data.	required
y	ndarray	Target data. Defaults to None.	None

Returns:

Type	Description
None	None

convert_scores_to_dataframe()

Convert the all_scores list to a DataFrame.

save_results()

Save optimization results and the best model to files.

Helper Functions

cdr_bench.optimization.optimization.perform_optimization(method, method_grid, method_param, X_transformed, y_transformed, scoring_params, dataset_output_dir)

Perform optimization using a specific dimensionality reduction method.

Parameters:

Name	Type	Description	Default
method	str	Dimensionality reduction method to use.	required
method_grid	Any	Parameter grid for the method.	required
method_param	Any	Parameters for the method.	required
X_transformed	ndarray	Transformed high-dimensional data.	required
y_transformed	Optional[ndarray]	Transformed reference data.	required
scoring_params	Any	Scoring parameters for optimization.	required

Returns:

Type	Description
Optimizer	Tuple[Any, np.ndarray, int]: Optimizer, coordinates, and nearest neighbors overlap score.

cdr_bench.optimization.optimization.create_param_grid(data_shape, n_components, method='UMAP', add_dim=False, test=False)

Create a parameter grid for the specified dimensionality reduction method.