Start hereWhat should you learn before fitting a model?
EDA is disciplined interrogation of the dataset: structure, distributions, relationships, missingness, anomalies and potential data-generating processes.
Building interactive view…
Technical lensFormalise what the visual is doing
Univariate summaries reveal centre/spread; bivariate views reveal association; multivariate analysis surfaces interactions and confounding. EDA must respect future validation boundaries when it influences modelling decisions.
Technical questionUse a tiny case to make the mechanism observable. Univariate summaries reveal centre/spread; bivariate views reveal association; multivariate analysis surfaces interactions and confounding. EDA must respect future validation boundaries when it influences modelling decisions. Verify one intermediate quantity, state change or mapping independently; then predict the consequence of this change: Hide an outlier or subgroup and see how a single aggregate can tell a different story.
Practitioner lensUse it responsibly
Ask a question before making a plot. Record observations, hypotheses and decisions instead of producing charts without interpretation.
Transfer testTransfer this idea to a new example and justify each decision using this practitioner rule: Ask a question before making a plot. Record observations, hypotheses and decisions instead of producing charts without interpretation. Then explain what should change if you deliberately test: Hide an outlier or subgroup and see how a single aggregate can tell a different story.
Worked explorationUse the visual as an experiment, not decoration
Use a small sales table. First inspect row grain, dtypes, missingness and duplicates. Then examine the sales distribution, compare regions, plot sales vs advertising, and inspect outliers. Record each observation as a question or data-quality finding rather than a causal conclusion.
Technical lens
Univariate summaries reveal centre/spread; bivariate views reveal association; multivariate analysis surfaces interactions and confounding. EDA must respect future validation boundaries when it influences modelling decisions.
Practitioner check
Ask a question before making a plot. Record observations, hypotheses and decisions instead of producing charts without interpretation.
Prediction before interactionHide an outlier or subgroup and see how a single aggregate can tell a different story.
Exploration walkthroughTurn the interaction into an evidence trail
Use a small sales table. First inspect row grain, dtypes, missingness and duplicates. Then examine the sales distribution, compare regions, plot sales vs advertising, and inspect outliers. Record each observation as a question or data-quality finding rather than a causal conclusion. Before moving the control, state your prediction. After the visual changes, name the specific state, statistic, boundary or mapping that changed and explain why that change is consistent—or inconsistent—with your prediction.
- Record one observable quantity before the interaction and the same quantity afterwards.
- Change one factor at a time so the causal effect of the control is inspectable.
- Use an edge or failure case to discover where the concept stops behaving as the simple story suggests.
