ruk·si
/notes

AutoML

Updated at 2018-07-06 20:52

Automated machine learning (AutoML) aims to abstract away most steps in machine learning workflow.

YMMV.

Usually, the system using AutoML will just try different algorithms and chooses which of the models gives the best predictions. But the system can also use meta-learning approaches.

graph TD
  subgraph TRADITIONAL
    A[Data Collection] --> B[Exploration]
    B --> C[Data Preparation]
    C --> D[Feature Engineering]
    D --> E[Modeling]
    E --> F[Training]
    F --> G[Hyperparameter Tuning]
    G --> H[Inference]
  end

  subgraph AUTOML
    I[Data Collection] --> J[Inference]
  end

Using AutoML solutions is awesome if it works for your problem. Most machine learning problems will require you to get your hands dirty though.

Tools and systems that have AutoML features: