Operator catalog: choose a job, then inspect the method

An operator is a numerical operation such as SNV, a learner such as PLS, or a sampling rule. A node tells the pipeline how to use it: transform features, transform targets, create training copies, tag observations or fit a model. The same transformer under sample augmentation and ordinary preprocessing has different effects on the dataset.

Start from the job you need. Each link below leads to the parameter table and explanation; the corresponding node links lead to a worked recipe, expected result figure, exercise and common mistake.

Job	Detailed method list	Workflow example
Correct spectra or change features	Transforms Reference	Preprocessing: change what the model sees
Scale or discretize the target	Transforms Reference	y_processing: learn on one scale, report on another
Predict a number or category	Built-in Models Reference	model: turn features into predictions
Assign independent validation units	Splitters Reference	split: define what must be unseen
Diagnose unusual observations	Filters Reference	tag: flag a sample and keep it
Exclude eligible training observations	Filters Reference	exclude: choose which training rows influence the fit
Generate plausible training variants	Augmentation Operators Reference	sample_augmentation: give training examples plausible variants
Build multiple representations	Transforms Reference	feature_augmentation: retain several views of the same observations and concat_transform: give one model several feature blocks
Combine sources or fitted predictions	Built-in Models Reference	branch: let the same samples follow different paths and merge: decide what to bring back together
Correct a base prediction	Built-in Models Reference	residual: learn an additive correction
Generate alternative recipes	Generator keywords: choices, combinations and their exact outputs	Generators: turn an experiment plan into concrete recipes
Inspect spectra, folds, copies or exclusions	Chart nodes: ask a concrete question of the current data	Visualization

Enumerated spectroscopy methods

These are the Python SDK’s maintained spectroscopy families, not UI component counts. A class listed here does not imply native/WASM execution support. For native IDs, supported host/runtime combinations and portable recipes, consult 9. Choose transforms and models for a reason and 10. Run the same recipe in your language.

Filters

Augmentations

GaussianAdditiveNoise · MultiplicativeNoise · SpikeNoise · HeteroscedasticNoiseAugmenter · LinearBaselineDrift · PolynomialBaselineDrift · WavelengthShift · WavelengthStretch · LocalWavelengthWarp · SmoothMagnitudeWarp · BandPerturbation · GaussianSmoothingJitter · UnsharpSpectralMask · BandMasking · ChannelDropout · LocalClipping · MixupAugmenter · LocalMixupAugmenter · ScatterSimulationMSC · PathLengthAugmenter · BatchEffectAugmenter · InstrumentalBroadeningAugmenter · DeadBandAugmenter · TemperatureAugmenter · MoistureAugmenter · ParticleSizeAugmenter · EMSCDistortionAugmenter · DetectorRollOffAugmenter · StrayLightAugmenter · EdgeCurvatureAugmenter · TruncatedPeakAugmenter · EdgeArtifactsAugmenter · Spline_Smoothing · Spline_X_Perturbations · Spline_Y_Perturbations · Spline_X_Simplification · Spline_Curve_Simplification · Rotate_Translate · Random_X_Operation

Compatible host estimators beyond this list

The Python SDK also accepts compatible sklearn estimators and installed neural frameworks. For regression, useful comparison families include Ridge/Lasso/ElasticNet, PLS, SVR, random forests, boosting, Gaussian processes and a dummy baseline. Classification counterparts include logistic regression, discriminant analysis, PLSDA, SVC, tree ensembles and calibrated probability models. Their input and target constraints still apply; discovery alone does not establish that an estimator is suitable.

Examples connect these choices to experiments: multiple models, PLS variants, augmentation, and CV strategies. Exact constructors, fitting methods and learned attributes are under API Reference.

How to use a catalog entry

Identify the measurement or prediction question.
Open the family reference and compare mechanisms and parameters.
Open the node page to understand shape, sample count, fitting population and prediction behavior.
Choose your language’s supported recipe from 10. Run the same recipe in your language.
Run the matching example and inspect its expected artifacts before composing a larger pipeline.

For typed sources, missing measurements, early/late fusion and target masks, use 4. Describe your measurements and align your sources and Public execution matrix. Preserve sample identity and source schema throughout.

Operator catalog: choose a job, then inspect the method

Enumerated spectroscopy methods

Transforms

Models

Splitters

Filters

Augmentations

Compatible host estimators beyond this list

How to use a catalog entry