SETS

`SETSCF(model, data, backend, mode, min_shapelet_len, max_shapelet_len, num_candidates_to_sample_per_case=20, time_contract_in_mins_per_dim=10, predefined_ig_rejection_level=0.001, max_shapelets_to_store_per_class=30, random_state=42, remove_self_similar=True, silent=False, fit_shapelets=True)`

Bases: CF

Calculates and Visualizes Counterfactuals for Multivariate Time Series in accordance to the paper [1]. The shapelet transofor adapted by [1] is based on prior work of [2].

References

[1] Bahri, Omar, et al. "Shapelet-based Temporal Association Rule Mining for Multivariate Time Series Classification". SIGKDD 2022 Workshop on Mining and Learning from Time Series (MiLeTS)" [2] ostrom, Aaron and Bagnall, Anthony}, "Binary shapelet transform for multiclass time series". Bostrom, Aaron and Bagnall, Anthony

PARAMETER	DESCRIPTION
`model`	Model to be interpreted. TYPE: `[Module, Callable, model]`
`dataset`	Reference Dataset of training and test data.
`backend`	desired Model Backend ('PYT', 'TF', 'SK'). TYPE: `str`
`mode`	Name of second dimension: `time` -> `(-1, time, feature)` or `feat` -> `(-1, feature, time)` TYPE: `str`
`min_shapelet_len`	Value for min length of extracted shapelets / must be greater than 0 TYPE: `int`
`max_shapelet_len`	Value for max length of extracted shapelets < timeseries must be smaller or equal than timeseries length TYPE: `int`
`num_candidates_to_sample_per_case`	number of assesed candiates per series TYPE: `int` DEFAULT: `20`
`time_contract_in_mins_per_dim`	Max time for shapelet extraction per dimension TYPE: `int` DEFAULT: `10`
`predefined_ig_rejection_levl`	Min Information Gain of candidate shapelet to keep TYPE: `float`
`random_state`	RandomState used throughout the shapelet transform TYPE: `int` DEFAULT: `42`
`remove_self_similar`	removes similar shapelets from a timeseries TYPE: `boolean` DEFAULT: `True`
`initial_num_shapelets_per_case`	Initial number of shapelets per case. TYPE: `int`
`silent`	logging. TYPE: `bool` DEFAULT: `False`

`explain(x, orig_class=None, target=None)`

Calculates the Counterfactual according to Ates. Arguments: x (np.array): The instance to explain. Shape : mode = time -> (1,time, feat) or mode = time -> (1,feat, time) target int: target class. If no target class is given, the class with the secon heighest classification probability is selected.

RETURNS	DESCRIPTION
`([array], int)`	Tuple of Counterfactual and Label. Shape of CF : `mode = time` -> `(time, feat)` or `mode = time` -> `(feat, time)`

`fit(occlusion_threshhold=0.1, remove_multiclass_shapelets=True)`

Calculates the occurences of shapelets and removes shapelets belonging to more than one class. This process can be triggered with different parameter options without a new shapelet transform run. Arguments: x (np.array): The instance to explain. Shape : mode = time -> (1,time, feat) or mode = time -> (1,feat, time) target int: target class. If no target class is given, the class with the secon heighest classification probability is selected.

RETURNS	DESCRIPTION
`([array], int)`	Tuple of Counterfactual and Label. Shape of CF : `mode = time` -> `(time, feat)` or `mode = time` -> `(feat, time)`