Start with your syllabus and your available time

Data science covers several connected areas: statistics, data preparation, programming, machine learning and communicating results. The balance varies between courses and roles, so begin with your own syllabus, lecture notes, assignments and any official assessment guidance. Do not assume that a plan built for one course matches another.

This four-week schedule uses four study sessions each week, usually 75–90 minutes each. That is about five to six hours weekly. If you have less time, keep the sequence and shorten sessions; if you have more, spend the extra time solving problems or applying methods to a dataset rather than adding more passive reading.

For each session, keep a specific question in view: for example, “When is a stratified split useful?” Work from your own material, attempt an answer, then check it against your notes. Record errors and uncertainty by topic. Those records—not how familiar a chapter feels—should determine what you revisit.

In MySummaries, a due queue can show which board sections need retrieval practice before you begin new work:

Due today
Statistics and probability8 due
Data preparation5 due
Model evaluationClear

An illustrative due queue for three data science boards.

Set up the boards in week one

Make a revision board for each major unit in your syllabus, or use one board with clearly separated sections if that suits your course. Import your lecture slides, PDFs, assignment feedback and photographs of handwritten notes. Check that important equations and code snippets have been captured correctly; OCR can misread symbols, subscripts and indentation.

Use a small number of sections that reflect what you need to retrieve and apply. For example, keep probability separate from model evaluation. A section should contain definitions, assumptions, formulas and worked examples from your own materials—not just a list of topic names. Mark a section for review when you have made cards from it or have a specific gap to address.

A board on this topic ends up looking like this:

Data Science Core methodsStudy
Foundations, preparation and modellingCore methods5 sections · 3 columns
Probability and inference4 due
  • Conditional probability: P(A|B) = P(A∩B)/P(B), for P(B)>0
  • Confidence interval: describes uncertainty in an estimated parameter under the stated method and assumptions
Classification metrics2 due
  • Precision: TP/(TP+FP); useful when false positives carry a substantial cost
  • Recall: TP/(TP+FN); useful when missing positive cases carries a substantial cost
  • Accuracy can mislead when class frequencies are strongly imbalanced
Data preparation3 due
  • Fit imputation, scaling and encoding steps on training data only
  • Use a pipeline so the same transformations are applied consistently at evaluation and prediction
Validation and leakage

Keep a final test set separate from choices made during training. Use cross-validation on training data for model selection; do not tune against the test set.

Regression
MethodTypical use or limitation
Linear regressionContinuous target; models a linear relationship in the predictors
Ridge regressionL2 penalty; can shrink correlated coefficients
Logistic regressionClassification model; outputs class probabilities after a link function

An illustrative board with sections for concepts, preparation, models and evaluation.

Week 1: establish the baseline

Session 1 — 75 minutes: map the syllabus. List the topics and learning outcomes. Match each to a source in your notes, and mark topics you cannot yet explain. Pick a small dataset from your course materials for later examples; do not spend the session searching for a perfect dataset.

Session 2 — 90 minutes: probability and descriptive statistics. Retrieve key definitions from memory, then work through examples involving conditional probability, mean and variance, distributions and sampling. For each formula, write what its terms mean and the conditions under which it applies.

Session 3 — 75 minutes: inference. Review the course’s treatment of estimation, confidence intervals and hypothesis tests. Practise interpreting a result in context. Avoid turning a p-value into the probability that the null hypothesis is true; check your course’s precise definitions and conventions.

Session 4 — 90 minutes: programming and data inspection. Use a notebook or equivalent environment to load a dataset, inspect types and missing values, summarise numeric columns and check category frequencies. Write down what each operation reveals. Finish by making a short list of uncertainties for week two.

Week 2: make the data and modelling choices explicit

Session 1 — 75 minutes: cleaning and missingness. From your notes, compare deletion, simple imputation and any methods taught in your course. Explain what each assumes or risks. Practise detecting duplicates, inconsistent categories and impossible values without altering the raw source.

Session 2 — 90 minutes: features and transformations. Work through scaling, categorical encoding and feature construction covered in your syllabus. Identify which methods are sensitive to feature scale. Practise putting transformations inside a pipeline, fitted on training data rather than on the complete dataset.

Session 3 — 90 minutes: regression. Fit or interpret a baseline linear model, then compare it with a regularised model if included in your course. Explain coefficients in context and distinguish prediction from causal interpretation. Check assumptions and diagnostics using the approach your course teaches.

Session 4 — 75 minutes: classification. Review logistic regression and the other classifiers in your syllabus. For each, state the target type, useful assumptions or limitations, and how you would produce predictions. Use a confusion matrix to connect false positives and false negatives to the problem being solved.

Week 3: test whether the model is useful

For a new lecture, choose a section where retrieval practice and written work both show difficulty. A picker might surface the topic that is costing you marks:

New lecture
How long have you got?5 min10 min15 min20 min
Model evaluationData preparationRegression

Model evaluation · Struggling — getting 4 of 9 cards wrong, scored 41% in your last practice set

Generate lecture ideas

An illustrative lecture picker highlighting a topic for focused review.

Use a short audio review to revisit the definitions and decision points, then spend the rest of the session solving a problem without the recording. Listening alone is not evidence that you can select a metric or explain a validation design.

Session 1 — 90 minutes: splits and validation. Explain why training, validation and test data have different roles. Practise identifying leakage in examples, including preprocessing performed before a split and repeated decisions based on test-set performance. Review how your course handles time-dependent data, where random splitting may not be appropriate.

Session 2 — 90 minutes: metrics and thresholds. Calculate accuracy, precision, recall and F1 from a small confusion matrix. Compare them for an imbalanced dataset, and state which error matters most in the stated context. If your course covers ROC-AUC or PR-AUC, explain what each summarises and its limitations; do not treat a metric as an automatic choice independent of the task.

Session 3 — 75 minutes: tuning and comparison. Review the model-selection procedure taught in your course. Practise comparing a simple baseline with a more complex model using the same validation approach. Note the metric, data split and decision rule so that the comparison is reproducible.

Session 4 — 90 minutes: timed mixed practice. Answer a set of questions across weeks one to three without notes. Include at least one interpretation of a model result and one question about data leakage. Mark each answer against your notes, and label the cause of each error: knowledge gap, calculation, method choice or unclear explanation.

Week 4: retrieve, explain and repair

Session 1 — 75 minutes: revisit the weakest areas. Rank your sections by practice accuracy and repeat errors, not by how much time you have already spent on them. Redo missed problems before reviewing the worked answer. If you still miss the same point, make one concise card phrased as a question with a single answer.

Session 2 — 90 minutes: end-to-end workflow. Take a dataset used in your course and describe the workflow from question definition to reporting. Include a baseline, a defensible split, preprocessing fitted on training data, model selection, appropriate evaluation and limitations. Use the requirements and terminology in your own course materials.

Session 3 — 75 minutes: communicate results. Practise explaining a model and its evaluation to someone who needs to make a decision. State the target, data used, metric, main result and a limitation. Avoid claims that go beyond the evidence, especially causal claims from a predictive analysis.

Session 4 — 90 minutes: final mixed review. Complete a timed set drawn from all four weeks. Mark it, then spend the final part of the session correcting the two most consequential misunderstandings. Keep a brief next-step list for any topics your syllabus still requires; do not try to replace an unstudied unit with a last-minute skim.

A weak-topic report should rank areas by your results in practice, not claim to predict what a particular course assessment will contain:

Where marks go missing
41%Model evaluation6×
63%Probability and inference5×
88%Data preparation7×

An illustrative report ranking practice sections by accuracy and how often they appeared in your sets.

How MySummaries helps

MySummaries lets you build revision boards from your own course materials, then turn board content into spaced-repetition cards, written practice and examiner-voice audio. For this plan, use the board to organise your syllabus, the due queue to schedule retrieval, and the weak-topic view to decide what to repair next.