The modelling half of an ML loop: supervised and unsupervised methods, features and leakage, metric choice, imbalance, deep learning foundations, forecasting and ranking.
24 published across 8 topics.
Linear and logistic models, trees and gradient boosting, regularisation, and choosing a family from the data and the constraints rather than by habit.
Start with gradient-boosted trees and keep a regularised linear model as the baseline. Boosting dominates business tables for structural reasons, not fashionable ones. It ignores feature scale, splits on thresholds rather than slopes, absorbs missing values, and finds interactions you never named.
Rank the causes by how often they are to blame and test them in that order: target leakage, a validation split that ignored group or time structure, train-serve skew, then genuine distribution shift. One cheap experiment separates most of them, which is rescoring logged production rows offline with the same artefact.
Regularisation adds a penalty on coefficient size to the objective, so each coefficient must buy its own weight with error. L2 shrinks smoothly and spreads weight across correlated features; L1 has a constant gradient towards zero, so it drives coefficients exactly to zero and selects features as a side effect.
Encoding, scaling and interaction terms, target and temporal leakage, train-serve skew, and the features that quietly encode the label.
One-hot is unusable at that width, so the options are frequency encoding, smoothed target encoding, hashing, or a library's native categorical handling. Target encoding is the strongest and the most dangerous, because it must be fitted inside each fold or it leaks the label. Use this categorical encoding answer to show the decision, trade-off, and evidence rather than a memorised definition.
Train-serve skew is a mismatch between features used in training and features produced at serving time. Prevent it with shared transformations, point-in-time joins, feature logging, schema checks and monitoring of served feature distributions. It also connects feature store to the point an interviewer is testing.
Target leakage happens when a feature contains information that would not be available at prediction time. Catch it by auditing feature creation times, window boundaries, post-label columns and suspiciously strong validation performance. Use this data leakage answer to show the decision, trade-off, and evidence rather than a memorised definition.
Cross-validation that respects structure, ROC versus precision-recall, calibration, threshold selection, and tying a metric to a business decision.
Build the split to imitate the gap between training and deployment. Group all of an entity's rows onto one side when rows repeat per entity, split forward in time for anything temporal with an embargo covering the label window, and treat a plain random shuffle as the default that silently inflates almost every score.
ROC-AUC vs precision-recall is a metric choice for imbalanced classification. ROC-AUC measures ranking across all negatives, while PR curves expose alert precision and recall at the positive-class prevalence.
A model is calibrated when among all cases it scores 0.30, close to 30% are positive. Most classifiers emit a score that ranks well and is not a probability, so any decision built from expected value is wrong by an amount nobody measures. Fit isotonic or Platt scaling on a held-out set to repair it.
Fraud and failure-scale imbalance, resampling and class weights, why accuracy is meaningless here, and evaluating at the operating point you will run.
Work backwards from the review capacity. Fix the alert budget the team can clear, evaluate precision and recall at exactly that budget on data at the real prevalence, define the label and its arrival delay before modelling, and never report accuracy, which a do-nothing model scores 99.95% on.
Choose a classifier decision threshold from calibrated probability, false-positive cost, false-negative cost and operational capacity. The threshold changes when costs, prevalence, calibration or alert budget changes. It also connects cost sensitive learning to the point an interviewer is testing.
Class weights, oversampling, undersampling and SMOTE are ways to train on imbalanced data, but none creates new signal. They change loss, sample frequency or geometry, so validation, calibration and threshold choice must be handled carefully. Use this class imbalance answer to show the decision, trade-off, and evidence rather than a memorised definition.
Clustering and its evaluation, dimensionality reduction, density and isolation methods, and detecting anomalies with no labels to learn from.
Model what normal looks like and score deviation from it, using distance or density methods, an isolation forest, or reconstruction error from an autoencoder. Then set the threshold from how many alerts the team can investigate per day, because an unlabelled score distribution cannot tell you where to cut.
Combine internal geometry measures such as silhouette with stability under resampling, then accept that neither picks k for you. A clustering is good when the segments are separable, reproducible on a resampled dataset, and distinct on a variable the business will act on. It also connects unsupervised learning to the point an interviewer is testing.
Dimensionality reduction helps when features are redundant, noisy or too expensive to serve, but it can hide the signal. PCA keeps variance, not relevance, and t-SNE or UMAP plots should guide exploration rather than prove model quality.
Backpropagation, optimisers and learning-rate behaviour, normalisation and regularisation, attention and transformers, and when a neural net is not the answer.
Backpropagation applies the chain rule backwards through a network, reusing each layer's cached activations to get every parameter's gradient in one pass. Gradients vanish when many small factors multiply, and it shows as a stalled loss with early-layer weights barely moving while the last layers still train. It also connects vanishing gradient to the point an interviewer is testing.
Choosing an optimizer and learning rate schedule starts with the learning rate, then the optimizer. AdamW is a strong default, SGD with momentum can generalise well when tuned, and warmup plus decay usually matters more than the optimizer name. Use this optimisers answer to show the decision, trade-off, and evidence rather than a memorised definition.
On small and medium tabular data, where gradient-boosted trees usually match it for a fraction of the effort; where a decision must be explained exactly rather than approximately; and where a tight CPU latency or memory budget makes it unservable. The simpler model then wins on every axis that matters.
Stationarity and seasonality, backtesting without lookahead, horizon and hierarchy, and why a random split ruins a forecasting model.
Forecast at whichever levels have signal, then reconcile so the numbers are coherent by construction. Bottom-up preserves detail but accumulates noise, top-down is stable but allocates by an assumed proportion, and optimal reconciliation projects independent forecasts onto the set that adds up.
Plot the series first, then use a decomposition and an ADF or KPSS test to separate trend from seasonality from noise. Difference or transform away a trend, use Fourier terms when several seasonal periods coexist, and treat holidays and calendar structure as features rather than outliers.
A random train-test split ruins a forecasting model by leaking future observations into training. Use time-based holdouts or rolling-origin backtesting, and build every feature only from data available before the forecast timestamp. Use this time series answer to show the decision, trade-off, and evidence rather than a memorised definition.
Collaborative and content signals, candidate generation then ranking, cold start, position bias and feedback loops, and offline-online metric divergence.
Because you cannot score a million-item catalogue per request inside a few tens of milliseconds. Retrieval reduces millions to hundreds with a cheap model judged on recall, then a far more expensive ranker orders those hundreds and is judged on the ordering it produces at the top of the list.
Fall back to content features and segment popularity while behavioural signal is absent, then blend towards collaborative scores as interactions accumulate. New items also need guaranteed exploration impressions, or the already-popular items take all the traffic and nothing can dislodge them.
Almost always the offline evaluation was measured on logs the old ranker produced. Clicks are confounded with position, the training data only contains items the old system chose to show, and a candidate that reproduces the old ordering scores well offline while adding nothing live.