[QUESTION] How to extract the Shap 'Explanation' object to use with the original Shap package?

[DataScientistET]

Is there a way to extract the Shap object so that I can use the original SHAP package to plot diagrams which are not yet implemented in darts? For example, I would like to plot but it requires the 'Explanation' object:

[madtoinou]

Hi @ettan10,

Due to the "temporal" aspect of the forecasts, the Explanation are stored in a dictionary;

from darts.datasets import AirPassengersDataset
from darts.explainability.shap_explainer import ShapExplainer
from darts.models import LinearRegressionModel

series = AirPassengersDataset().load()

model = LinearRegressionModel(lags=12)
model.fit(series[:-36])

shap_explain = ShapExplainer(model)
explanations = shap_explain.summary_plot()
# 1 for the horizon, "#Passengers" for the component
type(explanations[1]["#Passengers"])
>>> shap._explanation.Explanation

[DataScientistET]

Thanks @madtoinou. also can I confirm that the get_explanation function always returns an index relative to t0 timestamp regardless of what horizon parameter is passed to it? So if

• my data granularity is 1 hour
• set get_explanation(horizon=3)
• first index is = 2024-01-01 01:00:00

The actual forecast timestamp those SHAP values belong to in that row is for 2024-01-01 03:00:00 (i.e. for t=2)

[DataScientistET]

Also, how do I get the 'Explanation' object for an unseen test set since it is returned by the summary_plot() method.

For example, with your example above,

from darts.datasets import AirPassengersDataset
from darts.explainability.shap_explainer import ShapExplainer
from darts.models import LinearRegressionModel

series = AirPassengersDataset().load()

model = LinearRegressionModel(lags=12)
model.fit(series[:-36])

shap_explain = ShapExplainer(model)
shap_explain.get_explanation(foreground_series = unseen_series)

#how to get the explanation object from here?