Dataiku ML Practitioner study plan

Map the visual machine learning workflow end to end: identify where dataset preparation, feature handling, model training, evaluation, and deployment sit in the tool's interface.

Configure a project with a train/test split and a target variable, then explain in writing why the chosen split strategy fits the data.

Train at least three baseline algorithms (for example a regularized linear model, a tree ensemble, and a gradient boosting model) on the same dataset and compare their default metrics.

Record which preprocessing steps (missing value handling, encoding, scaling) each algorithm applies automatically versus which require manual configuration.

Diagnose overfitting and underfitting by training the same model at several complexity settings and plotting training versus validation performance.

Configure cross-validation for a model and compare its score against a single holdout split, explaining the difference in variability.

Interpret per-feature importance and partial dependence outputs for a tree-based model, then write what each tells you and what it does not.

Tune at least two hyperparameters using a search strategy and justify the search space you chose.

Select evaluation metrics appropriate to the problem type and class balance, and compute them on the same model to show how the ranking of models can change.

Build a confusion matrix and threshold analysis for a binary classifier, then pick an operating threshold with a stated business rationale.

Handle class imbalance using at least two techniques and compare their effect on the chosen metric.

Detect and address data leakage by auditing the feature set for target-derived or time-violating columns.

Deploy a trained model to a scoring endpoint and score a fresh dataset through it.

Set up a retraining or evaluation workflow that recomputes model metrics on new labeled data.

Document the model's inputs, outputs, assumptions, and known limitations in a model card.

Review the full visual ML domain and list the three subtopics you understand least.

Compute and interpret descriptive statistics for numeric and categorical variables, including distribution shape and outlier flags.

Build univariate visualizations for at least four variable types and state which question each chart answers.

Distinguish mean, median, and trimmed statistics on a skewed variable and explain when each is the right summary.

Identify missing data patterns and quantify the share of missingness per column.

Compute correlation coefficients across numeric variables and separate correlation from causation in written interpretation.

Run hypothesis tests for group differences and for association between categorical variables, then report effect size alongside p-values.

Check test assumptions and choose a nonparametric alternative when assumptions fail.

Interpret confidence intervals and explain what they do and do not say about a single sample.

Build a multivariate view combining at least three variables and describe interaction effects visible in the data.

Segment the dataset using grouping and aggregation, then compare segment-level statistics against the overall population.

Detect Simpson's paradox or a confounding pattern in your dataset or a constructed example, and explain the mechanism.

Create a chart that communicates a finding to a nontechnical audience and critique it for misleading scale or framing.

Re-run every weak-area exercise flagged in week 4 and in the statistics review, scoring each attempt against a written rubric.

Complete two timed mixed-domain practice sets covering both visual ML and interactive statistics under exam-like conditions.

Explain the connection between statistical testing and model evaluation metrics in a short written synthesis.

Consolidate a personal error log of recurring mistakes with the corrective rule for each.