scikit-survival
by K-Dense-AIscikit-survival skill for survival analysis and time-to-event modeling in Python. Use this guide for censored data, Cox models, random survival forests, gradient boosting, Survival SVMs, and survival metrics like concordance index and Brier score.
This skill scores 78/100 and is worth listing: it gives directory users a clearly triggerable, library-specific survival-analysis workflow with enough detail to justify installation, though it is not yet maximally operational. The score means it is a solid candidate for agents that need scikit-survival guidance, but users should expect some manual interpretation because the repo shows no helper scripts or supporting resources.
- Strong triggerability: the frontmatter explicitly says to use it for censored survival data, Cox models, Random Survival Forests, Gradient Boosting, Survival SVMs, and common survival metrics.
- Good operational scope: the body is substantial (14k+ chars) with many headings and workflow-oriented content, which suggests more than a stub or placeholder.
- Trustworthy listing signal: valid frontmatter, no placeholder markers, and repository/file references indicate a real skill page rather than a demo.
- No install command, scripts, or reference files are provided, so agents may need to infer setup and usage details from prose alone.
- Limited support scaffolding: the repo shows zero resources/rules/assets, which reduces progressive disclosure and makes edge-case adoption less predictable.
Overview of scikit-survival skill
The scikit-survival skill helps you work with survival analysis and time-to-event modeling in Python, especially when your data includes censoring and ordinary regression is not enough. It is best for analysts, data scientists, and ML practitioners who need to model event timing, compare risk across groups, or evaluate predictive survival models with metrics that respect censoring.
What makes the scikit-survival skill useful is its fit with the scikit-learn ecosystem: you can use familiar estimator-style workflows while applying survival-specific methods such as Cox models, random survival forests, gradient boosting, and survival SVMs. If you are deciding whether to install it, the main question is simple: do you need a practical scikit-survival guide for real censored outcomes, not just a generic explanation of survival analysis?
What this skill is for
Use it when the job is to predict time until an event, estimate risk over time, or compare survival curves from structured tabular data. It is a strong fit for clinical, reliability, churn, and other scikit-survival for Data Analysis use cases where event timing matters and some outcomes are incomplete.
Where it fits best
This skill fits best in Python workflows already using pandas, NumPy, and scikit-learn-style modeling. It is especially useful if you want to move from “I have time-to-event data” to a model, evaluation plan, and output you can explain.
Main adoption blockers
The biggest blockers are data preparation and metric choice: survival targets are not plain labels, and censoring must be represented correctly. If you are not ready to define event time, censoring status, and a sensible evaluation horizon, installation alone will not solve the problem.
How to Use scikit-survival skill
Install and open the right files
Install the scikit-survival skill with the directory’s normal skill install flow, then open SKILL.md first. Because this repository does not include helper scripts or extra reference folders, the main source of truth is the skill file itself plus any repository-wide conventions already in your environment.
Turn a rough goal into a usable request
A weak request says: “Analyze survival data.” A stronger request says: “Use scikit-survival to fit a Cox model on this right-censored dataset, compare it with a random survival forest, and report concordance index and time-dependent performance on a held-out set.” The more explicitly you name censoring, event definition, feature type, and evaluation metric, the better the output.
Inputs the skill needs
Provide:
- event type and censoring rule
- time column and event indicator
- feature columns and any exclusions
- target horizon or prediction use case
- preferred model family, if you have one
- constraints such as interpretability, calibration, or speed
If you are using scikit-survival usage in a notebook or codebase, also say whether you want code only, explanation only, or both.
Practical workflow
Start by asking for a data-shape check, then a model choice recommendation, then a training and evaluation plan. That order reduces avoidable errors, because survival modeling choices depend on whether the dataset is small, heavily censored, nonlinear, or meant for explanation rather than ranking.
scikit-survival skill FAQ
Is scikit-survival a good fit for beginners?
Yes, if you already know basic Python and some supervised learning. It is not beginner-friendly in the “guess and go” sense, because survival targets, censoring, and evaluation all need explicit handling.
How is this different from a normal prompt?
A normal prompt may describe survival analysis in general terms, but the scikit-survival skill is more useful when you need concrete implementation guidance: which estimator to use, how to encode outcomes, and how to evaluate predictions without ignoring censoring.
When should I not use it?
Do not use it if your problem is just binary classification, ordinary regression, or a non-time-based ranking task. It is also a poor fit if you do not know the event definition or cannot trust the time-at-risk information.
Does it fit the scikit-learn ecosystem?
Yes. That is one of its main advantages. If your workflow already depends on familiar estimator patterns, the scikit-survival install is a good choice because it aligns better with scikit-learn-style practice than a standalone survival tutorial.
How to Improve scikit-survival skill
Give the model the survival framing up front
The most useful inputs are the ones that remove ambiguity: what counts as the event, what is censored, and what prediction time matters. If you provide those three things, the skill can make much better choices about model form and evaluation.
Specify what “good” means
Tell the skill whether you care most about ranking risk, estimating survival probabilities, interpretability, or calibration. A Cox model and a random survival forest can both be valid, but they optimize different outcomes and produce different explanations.
Share data constraints and failure risks
Mention small sample size, heavy censoring, missing values, categorical encoding, class imbalance across events, or leakage risks from future information. Those details often matter more than the model family and help avoid misleading scikit-survival usage.
Iterate with one concrete output request
After the first answer, ask for one artifact at a time: a feature-prep checklist, a model comparison table, or code for fitting and scoring. That makes the scikit-survival guide more actionable and usually improves the next result faster than asking for “more detail” in general.