EDA Playbook

Demo

Click to play with sound.

---
name: EDA Playbook
description: Guides an analyst through a structured exploratory data analysis before any modeling begins. Apply when receiving a new dataset, debugging unexpected model behavior, or auditing data quality upstream.
---

# EDA Playbook

Skipping EDA before modeling is the single most common source of silent failures in ML pipelines. This skill establishes a repeatable, thorough checklist that surfaces problems early.

## 1. Shape and Schema Audit

Before anything else, verify the dataset dimensions and types match expectations.

- Confirm row count is in the expected range; flag if orders of magnitude off
- Check every column dtype; coerce-or-drop mismatches before downstream steps
- List columns with any null values and their null rates
- Flag columns with null rate above 20% for explicit handling decisions

## 2. Target Variable Analysis

Understand the label distribution before touching features.

- For classification: compute class frequencies and imbalance ratio
- For regression: plot the full distribution; check skew and kurtosis
- Identify any label values that are implausible or out of range… install to load the full skill