Ordinal Node Catalog
Ordinal category
Generated from 7 catalog nodes in AI/ML/Ordinal.
Nodes in this category
Showing 7 of 7 generated node docs.
Ordinal Metrics
AI/ML/OrdinalEvaluate predictions for an ordered target with distance-aware metrics. Plain accuracy is inadequate here: it treats "predicted high when the truth was medium" exactly as harshly as "predicted low", so a model that is reliably one level off scores like one that guesses. Quadratic weighted kappa is the standard headline metric because it weights every miss by how far off it was and corrects for chance agreement, but it answers only one of three questions: the linear kappa and the macro-averaged error say how far off the model is under a different cost structure and on the rare levels, while Kendall's tau-b and the Spearman correlation say whether it orders the rows correctly at all.
Train Ordinal Model (Adjacent Category)
AI/ML/OrdinalFit/Train an ordinal model that compares each level with the one directly below it: `log( P(level k+1) / P(level k) ) = contrast_k + x . beta`. Its coefficients answer `what does one more unit of this feature do to my rating?` - `exp(coefficient)` is the factor on the odds of scoring one level higher rather than staying put, the same factor at every step. That is NOT what Train Ordinal Model (Proportional Odds) reports: a cumulative coefficient is the log odds ratio of everything AT OR BELOW a cut point against everything above it, pooling levels instead of comparing two neighbours. The same fitted number therefore means different things in the two families, and since one shared coefficient applies once per step here, the bottom-to-top effect is (levels - 1) times the per-step effect. Pick this for ratings, severity grades and Likert answers, where the question really is about one step; pick proportional odds when the question is about crossing a threshold (`does this case escalate past level 2?`). Fitted by penalized maximum likelihood over all levels jointly, so per-level probabilities are calibrated and the Predict node returns a confidence. Scale your features first with the Fit Feature Scaler node: this is a gradient fit, and unscaled columns make it converge slowly or not at all.
Train Ordinal Model (Continuation Ratio)
AI/ML/OrdinalFit/Train a continuation-ratio model on an ORDERED target that is really a process that can halt. It fits K-1 sub-models, where sub-model k answers `given this row reached level k, did it STOP there?`, so the model describes a progression through the levels instead of placing cut points on a latent scale. Reach for it when the levels are genuinely sequential and each one had to be passed to get to the next: escalation tiers, disease stages, how far a signup funnel got, how far an incident escalated before it was contained. Each sub-model carries its own coefficient vector, so nothing assumes proportional odds, and the per-level probabilities are exact by the chain rule rather than differences of two fits. The cost is strictness: because each sub-model is conditioned on having reached its level, EVERY level must occur in the training data, middle ones included. Scale your features first with the Fit Feature Scaler node: these are gradient fits, and unscaled columns make them converge slowly or not at all.
Train Ordinal Model (Frank & Hall)
AI/ML/OrdinalFit/Train an ordinal model by decomposition: the ordered target is cut K-1 times (`is the level above this cut?`) and each cut is handed to an ordinary binary classifier, with the predicted level read back as the number of cuts answered yes. This is the one ordinal trainer here that is not linear in the features, so reach for it when the boundary between levels bends in a way the Proportional Odds and Ridge trainers cannot follow. The price is that the K-1 sub-models are fitted independently: there is no single latent scale, no coefficient vector to read a direction off, and no calibrated per-level probabilities - use Proportional Odds when you need those. Every declared level must occur in the training data at the bottom and at the top of the ordering, otherwise a cut has only one class and cannot be fitted. A Random Forest base is the sturdiest choice and by far the costliest: each cut grows its own full forest, so training costs K-1 forests and the saved model carries every tree of every one of them.
Train Ordinal Model (Neural CORAL/CORN)
AI/ML/OrdinalFit/Train a NEURAL ordinal model on a target whose levels are ORDERED (1 < 2 < ... < 5, or low < medium < high). This is the only trainer in the catalog that is BOTH non-linear in the features AND yields calibrated, rank-consistent per-level probabilities: Frank & Hall is non-linear but votes with K-1 independent classifiers and therefore carries no probability model, while every other ordinal node here is linear in the features. A small network feeds one of two rank-consistent heads, CORAL or CORN, and both are built so that P(y > k) can never rise with k for ANY parameter values — so the level probabilities are non-negative and sum to 1 with nothing patched up afterwards. THE HONEST LIMIT: leave Hidden Layers EMPTY and CORAL becomes exactly Train Ordinal Model (Proportional Odds) with Loss = AllThreshold and Margin = Logistic, and CORN becomes exactly Train Ordinal Model (Continuation Ratio) — the same objective in the same parameters. The hidden layers are the entire contribution, so if your problem is linear in the features prefer those nodes: convex objective, no seed dependence, readable coefficients, better tested. Reach for this one when the level is genuinely not monotone in the features (a boundary that bends back on itself, which no linear ordinal model can represent at all). Two costs come with the network: it has far more parameters than a linear model and so needs far more rows — check the Architecture output — and the objective is not convex, so the Seed changes the fit. Scale your features first with the Fit Feature Scaler node; unscaled columns make this converge slowly or not at all.
Train Ordinal Model (Proportional Odds)
AI/ML/OrdinalFit/Train a proportional-odds model on a target whose levels are ORDERED (1 < 2 < ... < 5, or low < medium < high). Use this instead of a classifier, which treats the levels as unrelated names and so counts predicting `low` for `high` as no worse than predicting `medium`. Use it instead of a regressor, which treats the levels as real numbers and so invents distances the levels do not carry (`high` is not exactly twice `medium`). The model learns one coefficient vector plus ordered cut points, which keeps predictions monotone in the score and, under the default loss, yields calibrated per-level probabilities. Link Function, Loss and Margin widen it to the whole threshold-model family, up to support vector ordinal regression, while Free Features relaxes the shared coefficient into one slope per cut point. Scale your features first with the Fit Feature Scaler node: this is a gradient fit, and unscaled columns make it converge slowly or not at all.
Train Ordinal Model (Ridge)
AI/ML/OrdinalFit/Train an ordinal model the cheap way: ridge-regress the level rank on the features, then cut the score at thresholds learned from the training distribution instead of rounding it. Closed-form, so it stays fast exactly where the proportional-odds model gets expensive - many levels, many features, or when you just want a quick ordinal baseline to beat. It also degrades gracefully when the proportional-odds assumption does not hold. Unlike the proportional-odds model it yields no probabilities: you get the predicted level and the latent score behind it, nothing calibrated.