DSMA Admission study plan

Write 40 SQL queries against a self-built 5-table practice schema, covering INNER/LEFT joins, GROUP BY with HAVING, and subqueries in WHERE and FROM clauses

Normalize a messy flat CSV into 3NF by hand, documenting each table's primary key, foreign keys, and the anomaly each split removes

Implement the same 10 aggregation questions in both SQL and pandas, then diff the outputs row by row to confirm equivalence

Draft a personal SQL style sheet covering NULL handling, join order, and when to prefer CTEs over nested subqueries

Refactor 5 messy pandas scripts into chainable pipelines using method chaining, avoiding in-place mutation and SettingWithCopyWarning

Implement groupby-apply, merge with indicator=True, and pivot_table from memory on a 100k-row synthetic dataset, timing each

Compute by hand the mean, median, variance, and standard deviation of a 12-value sample, then verify with code and explain the n-1 denominator

Profile a dataset for missingness, duplicates, and dtype errors, and write a cleaning function that logs every row it drops or imputes

Derive Bayes' theorem from the definition of conditional probability and apply it to a 2x2 confusion-matrix scenario with concrete counts

Simulate the central limit theorem in code: draw 1,000 samples of size 30 from a heavily skewed distribution and plot the sampling distribution of the mean

Write window-function queries using ROW_NUMBER, RANK, and LAG to compute running totals and month-over-month change

Explain when a LEFT JOIN silently duplicates rows and demonstrate the fix with a pre-aggregated subquery

Solve 15 probability problems covering independence, complement, and expected value, recording each mistake and its cause

Design a 10-question sampling plan for a customer survey, stating the target population, frame, sampling method, and the specific bias each choice avoids

Identify consent, minimization, and purpose-limitation issues in 3 short fictional data-collection scenarios and propose a concrete remediation for each

Distinguish correlation from causation using a confounder example, and sketch the causal diagram that makes the confounder explicit

Map a fictional dataset's fields to sensitivity tiers (public, internal, PII, sensitive PII) and state a retention period per tier

Build a leakage-free scikit-learn Pipeline with ColumnTransformer, fit only on training folds, and prove no test data touches the scaler

Implement k-fold cross-validation by hand on a small array, then confirm your fold means match sklearn's cross_val_score

Train logistic regression, decision tree, and random forest on one dataset and compare them on a held-out set using accuracy, precision, recall, and ROC AUC

Diagnose overfitting by plotting training versus validation error across tree depth, and state the depth where the gap widens

Write a feature-engineering function that encodes categoricals and handles unseen categories at inference time

Explain in writing how an unfair training label becomes a biased prediction, using a constructed example with base rates that differ across groups

Compute disparate impact ratio and equal-opportunity difference on a synthetic classifier's outputs, and interpret each against a stated threshold

Apply regularization (L1 and L2) to the same model and describe the effect on coefficients and validation error

Document a model card for your week-5 model: intended use, training data provenance, known limitations, and monitoring plan

Choose the correct chart type for 6 stated analytical questions and justify each rejection of the alternative you considered

Rebuild one default matplotlib chart into a publication-quality version: direct labels, no chartjunk, accessible color palette, annotated takeaway

Write a 150-word executive summary of your model results that leads with the decision, not the method

Critique 2 misleading charts (truncated axis, dual axis, cherry-picked range) and state the precise distortion each introduces

Take 2 timed full-length practice sets under exam conditions and log every error by domain and error type (concept gap, misread, calculation, time pressure)

Re-drill only the top 3 weakest subtopics identified in the error log, writing out the underlying rule for each from memory

Rehearse explaining your capstone project aloud in 3 minutes, covering governance choices, statistical assumptions, code design, model validation, and the visual story

Assemble a one-page formula and syntax reference covering SQL joins and windows, pandas idioms, probability rules, and model metrics