Dataiku AutoML
Dataiku is a platform where data work is assembled as a visual flow of datasets and recipes, and Dataiku AutoML is the modelling step inside it: you point at a prepared dataset, name the column you want predicted, and the platform trains and ranks candidate models for you.
What a run does
Training happens in the Lab, attached to the dataset rather than buried in a notebook. A prediction task offers presets — quick prototypes, interpretable models, high performance, deep learning — and from there the platform handles:
- Feature handling — per-column imputation, rescaling, dummy encoding and text vectorisation, each visible and overridable.
- Feature generation — pairwise interactions and polynomial combinations, plus automatic feature reduction.
- Algorithms and tuning — scikit-learn, XGBoost, LightGBM, Keras and Spark MLlib backends searched by grid, random or Bayesian strategies under a time limit, with cross-validation.
- Comparison — a leaderboard of trained models with variable importance, partial dependence and subpopulation analysis.
Why teams use it
The appeal is less the search than its place in a governed flow. A model you like is deployed into the flow as a saved model, applied by a scoring recipe, retrained when upstream data changes, and pushed to an API node for real-time prediction — with the lineage back to the raw source intact. Every automated choice stays editable, which is what lets an analyst start a model and a data scientist take it further in the same project instead of rebuilding it in code.
Dataiku AutoML is the automated modelling step inside the Dataiku platform: from a visual Lab attached to a dataset you choose a target column, and the system handles feature preprocessing and generation, trains scikit-learn, XGBoost, Keras and Spark MLlib candidates under a tuning budget, and ranks them on a leaderboard with interpretability reports. Its real value is placement — the chosen model deploys straight into the governed data flow for scoring, retraining and API serving.
Dataiku AutoML is a feature of Dataiku, a workbench where teams build their data work as a visual chain of steps rather than as pages of code. When it is time to make predictions — which customers will cancel, which invoices will be paid late — you point at your table, say which column you want predicted, and the tool trains a batch of models and shows them side by side with scores. You pick one, and it slots straight back into the same chain to run on new data. Analysts and engineers can work in the same project, which is the real draw.