How-To Guides

Start with the Walkthrough — a progressive, Python-first tour that covers a minimal run, inspecting the results, preprocessing, split methods, feature selection, and target transforms, each with a copy-paste YAML equivalent.

The focused guides below go deeper on individual topics:

Guide	When to use it
Walkthrough	Start here — the end-to-end tour in Python (+ YAML)
Recommend Models	Let COREMat suggest models for your dataset
Preprocessing per Model	Apply different preprocessing policies to different models
Models: Level and Overrides	Select a model preset or fine-tune individual model settings
Predefined Folds	Fix the CV folds yourself, with or without a hold-out
Submission Mode	Predict on unlabelled test rows and generate submission.csv
Inference	Load a saved pipeline and predict on new data
Register a New Model	Add a custom estimator to the COREMat model registry
Use Trained Models Programmatically	Access pipeline steps directly and extract TabPFN embeddings
Inspect the Saved Run Objects	Explore obj_split.joblib, obj_preprocess.joblib, and the per-round pipelines