How do you compute SHAP at production scale?

hard

Answer

$(2^{N}\;\mathrm{feature}\;\mathrm{subsets})$
$\mathrm{exact}\; + \;O(\mathrm{TLD}^{2})$
$(2) DeepSHAP: gradient-based for NNs.$
$(3) KernelSHAP with sampling for model-agnostic.$
$(4) Precompute + cache global feature importance.$
$(5) Compute per-prediction only when requested (customer-facing explanation) or async.$
$(6) Sample requests + summary dashboard.$
$Alternative: SAGE for global, faster than SHAP.$

Check yourself — multiple choice

Random
TreeSHAP exact fast trees + DeepSHAP gradient NN + KernelSHAP sampling model-agnostic + precompute global + on-request only for individual + sample summary; SAGE faster global alternative
Not possible
Not real

SHAP at scale: TreeSHAP / DeepSHAP / sample + async / SAGE.

#infrastructure#monitoring

Practise MLOps & Data Quality

215 interview questions in this topic.

Read the lesson Flashcards Take the quiz

Related questions