Saltar al contenido
TakoAPI
Volver a skills

senior-data-scientist

por alirezarezvani

Data & Analytics
2.1kDescargas según clawskills.sh
4Estrellas
1Vistas
0 me gustaInicia sesión para dar me gusta a esta skill
—Aún sin valoraciones
Tu valoración

Resumen

World-class data science skill.

Descripción

Qué hace esta skill

Covers statistical modeling and experiment design for production ML systems. Handles A/B testing with sample sizing and significance testing, feature engineering pipelines with Scikit-learn and XGBoost, cross-validated model evaluation with SHAP explanations, and causal inference via difference-in-differences. Works in Python, R, and SQL.

Bundles experiment design, feature engineering, model evaluation, and causal inference into one skill, so analysis stays statistically rigorous from design through deployment.

Cuándo usarla

  • Sizing an A/B test before launching a new checkout flow
  • Building a churn prediction model with cross-validated evaluation
  • Estimating the causal effect of a policy change on revenue
  • Engineering lag and cyclical features from transaction timestamps
  • Tracking and comparing model runs across experiments in MLflow

Flujo de trabajo de ejemplo

ENTRADA

User asks: Design and analyze an A/B test for a new email subject line with a 10% baseline open rate and a 5% minimum detectable effect

AGENTE
  1. 1

    Calculate required sample size per variant using power analysis with the given baseline rate and MDE

  2. 2

    Output a pre-registration checklist covering primary metric, randomization level, run duration, and sample ratio mismatch check

  3. 3

    Apply Bonferroni correction if the user specifies multiple secondary metrics

  4. 4

    Run analyze_experiment() on collected results and return lift, p-value, and 95% confidence interval

SALIDA

Sample size estimate, experiment checklist, and analysis results with lift and confidence interval interpretation

Instalación

clawhub install alirezarezvani/senior-data-scientist

Ejecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.

Enlaces