Overview¶
IAML (Integrated AutoML for Medical Labs) is a Python framework for research teams working with tabular clinical data. It combines preprocessing, model search and evaluation for classification, regression and survival analysis.
IAML offers an integrated workflow with ready-to-use defaults and a pipeline API for teams that want to define their own methods. Both use the same search, cross-validation and final fitting engine.
Who is it for?¶
Clinical researchers and data scientists working with Python and pandas. Researchers define the study population, outcome, features and evaluation design; IAML searches prediction pipelines within that design. The selected methods and parameters remain available for review.
An integrated workflow, ready to use¶
Start with IAML().fit(X, y) to search preprocessing methods, predictors and
their parameters using the built-in pipeline. You do not need to construct
a pipeline or configure each step to obtain trained candidates.
Follow Quick Start to run a first analysis. The From data to explanations connects the same calls to performance figures, explanations and saved outputs.
A pipeline API for teams going further¶
Adapt the preset or compose a recipe from reusable components. Choose the permitted alternatives, edit parameter domains and configure the analyses around training. Your own components can join these recipes through the extension interfaces.
Customize IAML introduces these possibilities visually. Pipeline API directs you to the construction, study configuration and extension guides when you are ready to customize a workflow.
What does a run produce?¶
The search compares candidates by cross-validation. Each candidate combines
preprocessing, a predictor and its scores; retained candidates are then refitted
on the chosen training population, using all training rows by default.
search.chosen_candidate is the selected fitted
candidate, including the preprocessing needed for new observations.
The study continues through the following outputs:
Stage |
Output |
Guide |
|---|---|---|
Describe and train |
Cohort summaries on request, then fitted candidates and validation scores. |
|
Evaluate |
Held-out scores and task-specific performance figures. |
|
Explain |
Selected methods and feature contributions for predictions. |
|
Report |
Saved outputs, experiment settings and method references. |
Descriptive statistics, figures and explanations are calculated when requested.
They do not all run automatically during fit. See How it works for
how candidate generation, evaluation and optimization fit together.
Project and contributions¶
IAML is developed by IIAS. Source code and issue tracking are available in the IAML GitHub repository. Contributions, bug reports and examples of research workflows are welcome.