Explain ML predictions with SHAP: explainer selection, additivity checks, honest reporting
Rigorous SHAP playbook for fitted models — define the explanation target, select explainer + masker from a decision table, validate additivity, plot the question (not just the available plot), report limitations
- What
- Rigorous SHAP playbook for fitted models — define the explanation target, select explainer + masker from a decision table, validate additivity, plot the question (not just the available plot), report limitations
- Cost
- Free
- Needs
- Python 3.12+ with uv; a fitted model to explain — the skill is a guided methodology, not software to install
- Install
- Copy the installer prompt below into your Muse — your agent does the rest.
Curated by Skill Harbor — @k-dense-ai's SHAP skill for explaining fitted machine-learning models, aligned with SHAP 0.52.0 and the modern `shap.Explanation` API: a 7-step rigorous workflow. Step 1, define the explanation target (model version, exact callable, output name/index and units, evaluation rows, background population). Step 2, select explainer and masker from a decision table (TreeExplainer, LinearExplainer, ExactExplainer, PermutationExplainer, PartitionExplainer, Deep/GradientExplainer, legacy KernelExplainer — with per-family constraints). Step 3, compute a modern Explanation with a complete binary-classification worked example. Step 4, control tree output semantics (probability/log-loss spaces need interventional masking). Step 5, use model-agnostic callables deliberately with budgeted permutation evals. Step 6, "visualize the question, not merely the available plot" — a question-to-plot mapping (beeswarm, waterfall, scatter, heatmap, text/image). Step 7, report limitations with results: baseline, explainer, masker, additivity error, correlated features, and a clear non-causal statement. Strong integrity rules throughout: never use SHAP as a substitute for predictive validation, never silence an additivity failure, never load untrusted pickle/joblib artifacts (code execution risk). Honest caveats: requires Python 3.12+ and uv for SHAP 0.52.0; SHAP describes model behavior — it does not establish causality or fairness (a small protected-feature attribution does not rule out proxy discrimination); the skill asks that substantial use be cited in manuscripts (arXiv:2609.00065). MIT licensed. Skill Harbor never reviews the code, review it yourself before use. Discovered via skills.sh.
Version:
Install
Prerequisites: Python 3.12+ with uv; a fitted model to explain — the skill is a guided methodology, not software to install Install "Explain ML predictions with SHAP: explainer selection, additivity checks, honest reporting" for me. It gives my agent @k-dense-ai's SHAP playbook: define the explanation target precisely, select explainer and masker from the decision table, compute modern shap.Explanation objects, control tree output semantics, validate additivity (never silence failures), never load untrusted pickle/joblib artifacts, visualize the question, and report limitations honestly — with a deterministic bundled tabular_report.py example. MIT-licensed. Repository: https://github.com/k-dense-ai/scientific-agent-skills/blob/main/skills/shap/SKILL.md 1. Fetch the SKILL.md file (and any helper files) from the repository path into a temporary folder and summarize what it does in one or two sentences. 2. Safety check: review the SKILL.md and scripts for anything suspicious (unexpected network calls, shell commands, credential harvesting). This repo should contain zero secrets in code, credentials only via the secure vault, allowed hosts declared in the SKILL.md. Verify that holds here; STOP on any red flag and tell me. 3. Install it as a skill: copy SKILL.md and its helper files into the agent's skills directory, in a folder named "shap". 4. Verify with no network calls: frontmatter valid, files in place. 5. Report what was installed, where, and what I still need to do myself (e.g. set up the Python 3.12 env and install shap[plots]==0.52.0 via uv; point the agent at my fitted model and the rows to explain; cite the Scientific Agent Skills paper (arXiv:2609.00065) if it materially contributed to a manuscript). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.
Questions
How do I install a build?
Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the build up for you — no manual configuration.
Where does my money go?
Straight to the seller. Skill Harbor never processes payments: checkout happens on the seller’s own page, usually Stripe.
What does the ✓ next to a creator’s name mean?
It means we confirmed the identity of the person behind the listing. It says nothing about the code itself — always check a build before installing it.