Can Dynamic Double Machine Learning use SHAP?

When there is only one panel, it works fine, but multiple panels will report “list index out of range”.for exmaple

Define DGP parameters

np.random.seed(123)
n_panels = 5000 # number of panels
n_periods = 3 # number of time periods in each panel
n_treatments = 3 # number of treatments in each period
n_x = 100 # number of features + controls
s_x = 10 # number of controls (endogeneous variables)
s_t = 10 # treatment support size

Generate data

Define additional DGP parameters

het_strength = .5
het_inds = np.arange(n_x - n_treatments, n_x)

Generate data

dgp = DynamicPanelDGP(n_periods, n_treatments, n_x).create_instance(
s_x, hetero_strength=het_strength, hetero_inds=het_inds, random_seed=12)
Y, T, X, W, groups = dgp.observational_data(n_panels, s_t=s_t, random_seed=1)
ate_effect = dgp.true_effect
het_effect = dgp.true_hetero_effect[:, het_inds + 1]
#%%

est = DynamicDML(
model_y=RandomForestRegressor(),
model_t=RandomForestRegressor(),cv=3)
est.fit(Y, T, X=X, W=W, groups=groups, inference="auto")
#%%
import shap

explain the model's predictions using SHAP values

shap_values = est.shap_values(X, feature_names=['A', 'B', 'C'], background_samples=100)
shap.summary_plot(shap_values['Y0']['T0'])

Start at the DynamicDML.shap_values entry point and reproduce the reported example with one panel and multiple panels using the shown SHAP call. Trace the “list index out of range” failure in the multi-panel case; done means the call returns SHAP values that can be passed to shap.summary_plot for multiple panels.