Role Pack · Mid to Senior
Data Scientist Interview Pack
Statistics, ML fundamentals, model evaluation and the ML system-design round.
From bias-variance to feature engineering to evaluating and deploying a model.
- Questions
- 360
- Topics
- 20
- Pages
- 301pages
- 360 questions, every one with a model answer
- ML + stats cheat sheet
- Model evaluation metrics
- ML design walkthroughs
Get the Data Scientist Interview Pack — $9
One-off. Delivered as a PDF by email the moment payment clears.
Prices shown are a USD base. Your local-market price is calculated at checkout in your own currency, plus any local sales tax.
What is inside
20 chapters, in the order the interviews tend to ask them:
- Probability, conditional probability and Bayes' theorem
- Distributions, the central limit theorem and sampling
- Point estimates, bias, variance and confidence intervals
- Hypothesis tests, p-values, power and multiple comparisons
- The linear algebra behind the estimators
- Linear regression and its assumptions
- Generalised linear models: logistic, Poisson and beyond
- Regularisation, the bias-variance penalty and model selection
- Bias, variance and reading resampling results
- Decision trees, random forests and gradient boosting
- Feature engineering, encoding and leakage
- Imbalanced classes, thresholds and calibrated probabilities
- Clustering and dimensionality reduction
- Time series, stationarity and forecasting
- Bayesian inference, priors and MCMC
- Neural networks, gradients and optimisation
- Text, embeddings and transformer models
- SQL for analysis and data modelling
- Pipelines, data quality and reproducibility
- Machine learning system design and deployment
How the answers are built
Each chapter opens with the checklist the question is really testing. A few from this pack:
- Probability is a number in [0, 1] satisfying non-negativity, additivity over disjoint events, and normalisation over the whole sample space.
- A distribution is a normalised weight function; the name you pick is a modelling claim about the data generating process.
- An estimator is unbiased when its expected value equals the parameter, which is a property of the procedure rather than of one dataset.
- A p-value is the probability of data at least as extreme as observed, computed under the null, and is not the probability the null is true.
- A matrix is a linear map; a decomposition is a way of breaking that map into pieces you can reason about independently.
- Ordinary least squares minimises squared residuals, which makes it a maximum likelihood estimator under Gaussian noise.
Questions this pack covers
- How do the probability axioms constrain a model someone has proposed?
- What does it actually mean for two events to be independent?
- What does the central limit theorem actually guarantee about sample means?
- Why does averaging fail to tame a heavy tailed variable?
- What does it mean for an estimator to be unbiased?
- Why is a confidence interval not a statement about this particular parameter?
- What exactly does a p-value measure under the null hypothesis?
- Why does a small p-value not mean the null is false?
Each one is answered in full in the pack, with the reasoning written out rather than a bullet list.
About this pack
- How many questions?
- 360, spread across 20 topics, written out in full across 301 pages.
- Do the answers just list keywords?
- No. Each answer is written as prose, because that is what you have to reproduce when someone asks you the question.
- How do I get it?
- Pay once and the PDF arrives by email immediately, with a signed download link.
- Can I get it refunded?
- Yes, if it is not what you expected. See the refund policy.
