TSR

FeatureAttribution(mlmodel, mode)

Bases: InterpretabilityBase

Abstract class to implement custom interpretability methods

mlmodel:

Methods

explain: explains a instance Returns

---

None

explain() abstractmethod

Explains instance or model. Parameters

---

instance: np.array Not encoded and not normalised factual examples in two-dimensional shape (m, n). Returns

---

pd.DataFrame Encoded and normalised counterfactual examples.

plot(item, exp, figsize=(6.4, 4.8), heatmap=False, normelize_saliency=True, vmin=-1, vmax=1, save=None)

Plots expalantion on the explained Sample. Parameters

---

instance: np.array timeseries instance in two-dimensional shape (m, n). exp: expalantaion Returns

---

matplotlib.pyplot.Figure Visualization of Explanation

TODO CHeck with all typer

Saliency(model, NumTimeSteps, NumFeatures, method='GRAD', mode='time', normalize=True)

Bases: FeatureAttribution

Base Method for Saliency Calculation based on [1]. Please use the designated Subclasses SaliencyMethods_PYT.py for PyTorch explanations and SaliencyMethods_TF.py for Tensforflow explanations.

References

[1] Ismail, Aya Abdelsalam, et al. "Benchmarking deep learning interpretability in time series predictions." Advances in neural information processing systems 33 (2020): 6441-6452.

---

PARAMETER	DESCRIPTION
model	model to be explained. TYPE: [Module, models]
NumTimeSteps	number of timesteps. TYPE: int
NumFeatures	number of features. TYPE: int
method	Saliency Method to be used. TYPE: str DEFAULT: 'GRAD'
mode	Second dimension 'time'->(1,time,feat) or 'feat'->(1,feat,time). TYPE: str DEFAULT: 'time'
normalize	Wheather or not to normalize the results TYPE: bool DEFAULT: True

plot(item, exp, figsize=(6.4, 4.8), heatmap=False, save=None)

Plots explanation on the explained Sample.

PARAMETER	DESCRIPTION
item	instance to be explained,if mode = time->(1,time,feat) or mode = feat->(1,feat,time). TYPE: array
exp	explanation, ,if mode = time->(time,feat) or mode = feat->(feat,time). TYPE: array
figsize	desired size of plot. TYPE: (int, int) DEFAULT: (6.4, 4.8)
heatmap	'True' if only heatmap, otherwise 'False'. TYPE: bool DEFAULT: False
save	Path to save figure. TYPE: str DEFAULT: None