Pipeline API¶
Use the pipeline API when your team needs to adapt the AutoML preset, assemble reusable strategies or integrate its own components. Recipes describe the allowed pipelines and their parameter domains; IAML handles candidate generation, cross-validation and final fitting with the same engine as the integrated workflow.
If you are new to this part of IAML, start with Customize IAML for a visual tour, or the example catalogue for runnable Python examples. Pipeline configuration remains optional for training with the defaults.
Choose your next step¶
Goal |
Read |
|---|---|
Adapt the preset or build your own pipeline. |
Build and adapt a pipeline: composition, subtraction, editing, reusable fragments and parameter domains. |
Configure the calculations and search around it. |
Configure a complete study: objectives, validation settings, descriptive statistics, explanations and analytical reports. |
Find a method’s exact behavior. |
Pipeline and study API reference: signatures, naming rules, atomic edits and diagnostics. |
Run a complete example. |
Pipeline and study examples: thirteen runnable Python examples. |
Implement a component for your team. |
Extending IAML: transformation, prediction, analysis and other extension contracts. |
A recipe and a fitted model have different roles. IAML(pipeline=recipe)
copies the declaration; fit creates execution instances. Edit the attached
copy through search.pipeline to prepare another run. Trained Candidate
objects retain their own fitted IAMLPipeline and analysis definitions.