ClassOverlapExplorer
Explorer that measures how much the classes of a dataset overlap.
It reports three of the geometrical complexity measures surveyed in Lorena et al. (2019), computed from the data alone, with no model involved:
F1
Maximum Fisher discriminant ratio. Near 0 when at least one feature
separates the classes on its own, near 1 when none does.
N1
Fraction of points sitting on a class boundary, taken from a minimum
spanning tree over the sample. A large value means a long, populated
frontier between classes.
N2
Ratio between the distance to the nearest same-class neighbour and the
distance to the nearest different-class neighbour. A large value means
the classes interleave rather than forming compact groups.
All three are normalised so that lower means easier to separate.
Use this explorer before training to tell an unpromising dataset from an unpromising model: if the classes already overlap heavily, a disappointing accuracy is a property of the data rather than of the learner, and the useful next step is feature engineering or relabelling rather than model search.
Parameters
- target_column, default=
- Required. Name or index of the column holding the class labels. It is added to the exploration automatically, so it does not need to be selected as a column.
- max_samples : integer, default=
2000 - Maximum number of rows used for the distance-based measures N1 and N2. Larger datasets are subsampled keeping the class proportions.
- random_state, default=
0 - Seed used when the dataset has to be subsampled.
Methods
get_results(self, exploration_path: str, options: Dict[str, Any]) -> Dict[str, Any]
ClassOverlapExplorerLoad and return the saved measures table for the frontend.
Parameters
- exploration_path : str
- Path to the JSON file saved by
save_notebook. - options : Dict[str, Any]
- Rendering options from the frontend (unused).
Returns
- Dict[str, Any]
- Dictionary with keys
"data"(nested dict of the table, keyed by column then row, which is the orientation the artifact conversion expects),"type"("tabular") and"config".
launch_exploration(self, dataset: 'DashAIDataset', explorer_info: DashAI.back.dependencies.database.models.Explorer) -> Any
ClassOverlapExplorerCompute the complexity measures and lay them out as a table.
Parameters
- dataset : DashAIDataset
- Dataset holding the feature columns and the target column.
- explorer_info : Explorer
- The explorer database record (unused).
Returns
- Any
- A
pandas.DataFrameindexed by row label, with a"value"column and a"detail"column describing each entry.
prepare_dataset(self, loaded_dataset: 'DashAIDataset', columns: List[Dict[str, Any]]) -> 'DashAIDataset'
ClassOverlapExplorerExtend column selection to include the target column.
Parameters
- loaded_dataset : DashAIDataset
- The full dataset.
- columns : List[Dict[str, Any]]
- Explicitly selected column descriptors.
Returns
- DashAIDataset
- Dataset containing the selected feature columns plus the target.
save_notebook(self, notebook_info: DashAI.back.dependencies.database.models.Notebook, explorer_info: DashAI.back.dependencies.database.models.Explorer, save_path: 'Path', result: Any) -> str
ClassOverlapExplorerSave the measures table to a JSON file on disk.
Parameters
- notebook_info : Notebook
- The notebook database record (unused).
- explorer_info : Explorer
- The explorer record used for filename generation.
- save_path : Path
- Directory where the file will be saved.
- result : Any
- The
pandas.DataFramereturned bylaunch_exploration.
Returns
- str
- The path of the saved JSON file as a POSIX string.
get_credential(self, name: str)
ConfigObjectResolve a registered credential component by name.
Parameters
- name : str
- Credential component class name (e.g. "HuggingFaceCredential").
Returns
- BaseCredential
- An instance of the requested credential component.
get_metadata(cls) -> Dict[str, Any]
BaseExplorerGet metadata for the explorer, used by the DashAI frontend.
Returns
- Dict[str, Any]
- Dictionary containing display name, description, image preview path, category, icon, color, allowed semantic types, allowed dtypes, and input cardinality constraints.
get_schema(cls) -> dict
ConfigObjectGenerates the component related Json Schema.
Returns
- dict
- Dictionary representing the Json Schema of the component.
validate_and_transform(self, raw_data: dict) -> dict
ConfigObjectIt takes the data given by the user to initialize the model and returns it with all the objects that the model needs to work.
Parameters
- raw_data : dict
- A dictionary with the data provided by the user to initialize the model.
Returns
- dict
- A validated dictionary with the necessary objects.
validate_columns(cls, explorer_info: DashAI.back.dependencies.database.models.Explorer, column_spec: Dict[str, Dict[str, str]]) -> bool
BaseExplorerValidate that the selected columns satisfy the explorer's constraints.
Parameters
- explorer_info : Explorer
- The database record for the explorer instance, including the selected columns.
- column_spec : Dict[str, Dict[str, str]]
- A mapping from column name to a dict with at least
"type"(semantic type name) and"dtype"(dtype string).
Returns
- bool
- True if all column constraints are satisfied, False otherwise.
validate_parameters(cls, params: Dict[str, Any]) -> bool
BaseExplorerValidate explorer parameters against the explorer's schema.
Parameters
- params : Dict[str, Any]
- The configuration parameters to validate.
Returns
- BaseExplorerSchema
- The validated and parsed schema instance. Subclasses that override this method may return a bool to indicate pass/fail without returning the model instance.