Comparisons
Pairs that get conflated in real metric reviews and real design docs — batch and online, precision and recall, data drift and concept drift, validation and test. Neither column wins; what decides is the problem. Each record leads with the confusion, because the confusion is the reason the record exists.
k-fold cross-validation vs a single holdout
k-fold is taught as the rigorous option and the holdout as the lazy one, and on temporal or grouped data the opposite is true. Folding a time series shuffles the future into the training set of every fold; folding user rows puts each user on both sides of some fold; the averaged metric is then a precise estimate of the wrong quantity. The strongest form of the k-fold side is statistical: on a few thousand exchangeable rows, a single split's metric moves by more than the differences you are trying to detect, and averaging over folds is the only way to see whether a change is real. It also uses every row for validation once. The strongest form of the holdout side is that a single forward-in-time holdout simulates deployment — train on the past, evaluate on the future — which is the question you are actually asking, and that a large holdout has an error bar small enough not to need folding. Grouped and time-series variants of cross-validation exist and are the reconciliation: fold by group, or roll the origin forward. The choice is decided by how the model will be used, not by which one sounds more careful.
When the data is small enough that one split gives a noisy metric, when the rows are exchangeable — no time order, no shared entities — and when training k times is affordable.
When the data has an arrow — time — so that only a forward split is honest, when training is expensive, or when the dataset is large enough that one held-out set already gives a tight estimate.
| Dimension | k-fold — split the data into k parts, train k times, average the validation metric | Single holdout — one training set, one validation set, one number |
|---|---|---|
| Metric variance | Lower — averaged over k estimates | Higher on small data, fine on large |
| Training cost | k runs | One run |
| Temporal data | Leaks unless rolled forward in time | Honest if the split is by time |
| Grouped data | Leaks unless folded by group | Honest if the split is by group |
| Simulates deployment | No — every fold trains on some future | Yes, if forward in time |
| Best for | Small, exchangeable data; model selection | Large data; time-ordered data; final evaluation |
| Still needs | A test set never used for selection | A test set never used for selection |
Model families compared
Linear, trees, boosting, neural and k-NN — compared without naming a winner, and with the block that says where the comparison stops being true.
No column is a winner. Each family is a set of assumptions about the data — linear separability, axis-aligned interactions, additive residuals, a representation that can be learned, locality in feature space — and the right one is the one whose assumptions your data happens to satisfy at the size you have. The reflex answer “XGBoost” is the red flag this table exists to catch: it is often right on tabular data and it is never right as a reflex, because it skips the baseline that would have told you whether anything more than a linear model was needed. The where this comparison misleads block on every row is the part worth reading.
Count independent entities, not rows — a million events from ten thousand users is ten-thousand-sized data for generalising to new users. And the neural column flips completely with transfer learning: a pretrained model fine-tuned on two thousand images beats every other column on that task.
The "boosting wins on tabular" claim is true for a tuned model on tens of thousands of clean rows with a validation set. It is not true for three hundred rows, for a problem whose signal is linear, for a regulator who wants coefficients, or for a team with no time to tune — and the gap to the linear model is often inside the error bar.
The columns are not competing on the same input. Once a pretrained network has produced an embedding, a linear model or k-NN on top of it is often within a few points of full fine-tuning — so "neural for images" usually means "a neural representation, then whichever head is cheapest".
Interpretability is not one property. A linear model with two hundred correlated features and L1 selection is harder to explain honestly than a depth-three tree; and every column's "explanation" is undermined equally by a leaked or proxy feature, which the explanation will present with confidence.
Training cost is dominated by the number of runs, not the run: a boosted model tuned over two hundred configurations costs more than a network fine-tuned once. And the k-NN column's zero is a loan repaid on every query.
The model is rarely the slow part. Feature retrieval, a network hop and JSON serialisation usually dwarf any of these numbers, so a latency budget is a question about the serving path before it is a question about the family.
A low burden on the modelling side does not remove the burden on the serving side: every column's features must be reproduced at prediction time with the same code, the same freshness and the same point-in-time semantics. Trees remove the need to engineer features, not the need to serve them.
Native handling is convenient and dangerous in equal measure: it lets a tree model learn that "missing" predicts the target, which is fine until serving produces missingness for a different reason — a timeout, a new form — and the model reads the outage as a signal.
Calibration only matters if a downstream decision multiplies the score by a cost or compares it to a probability threshold; a pure ranker does not need it. And calibration measured on the validation set drifts with the base rate in production, for every column alike.
Neither behaviour is "correct". A tree that predicts last year's maximum for a record-breaking day is wrong; a linear model that predicts a negative price is wrong differently. The honest answer is a monitor on inputs outside the training range and a fallback for them, whichever family serves.
Irrelevant is not the danger — leaky is. Every column will seize a feature that carries the answer, and the more capable the model the more efficiently it does so; robustness to noise says nothing about robustness to leakage, which only a point-in-time audit provides.
Every column is wrong somewhere, and the question that finds where is the same for all of them: how much data, of what modality, at what latency, with what interpretability requirement, judged by which metric, at what cost to train and serve. A model family chosen before those are answered is a guess with a library name.