What is Automated Machine Learning (AutoML)?

machine learning

AutoML, short for Automated Machine Learning, is the automation of the machine learning process to make machine learning jobs simpler, easier, and faster. One might ask, “What’s being automated?” Let’s look at the typical machine learning model building process.

Machine Learning Model Building Process

  1. Find a business problem
  2. Translate the business problem into a data science problem
  3. Find the necessary dataset
  4. Define a goal and metrics to measure, in both building a machine learning model and using the model in business
  5. Explore dataset
  6. Building a model by doing (not necessarily in this order and not limited to),
    1. Feature engineering
    2. Feature selection
    3. Algorithm selection
    4. Hyperparameter optimization
    5. Stacking
    6. Ensembling
  7. Deploy the model
  8. Evaluate the model, go back to 6 if not satisfied
  9. Productionize the model (usually using the model in other business applications)
  10. Use the model

Out of the 10 steps, AutoML typically involves automating the tasks described in step #6. By automating step #6, the machine learning process is now simpler, easier, and faster. However, AutoML generates new problems. The very first one is, “How can I trust the outcome of a machine?” Many AutoML users ask “Is that model the really best model?” Therefore, many technologies offer the ability to tweak some hyperparameters to users so users can build different models for comparison. While it’s understandable, this now removes the advantages of AutoML being simple and fast. 

The second problem is that typical AutoML technology requires technical knowledge. In other words, it often requires a data scientist to review the built model, find the best model or models to deploy. If there are multiple models that performed well in the evaluating phase, the data scientist might try different techniques to build one final model to productionize. This also removes the advantages of AutoML being simple and fast.

Some experts are now separating the AutoML into levels, such as the following:

Level 0No automation. You code your own ML algorithms. From scratch. In C++.
Level 1Use of high-level algorithm APIs. Sklearn, Keras, Pandas, H2O, XGBoost, etc.
Level 2Automatic hyperparameter tuning and ensembling. Basic model selection.
Level 3Automatic (technical) feature engineering and feature selection, technical data augmentation, GUI.
Level 4Automatic domain and problem-specific feature engineering, data augmentation, and data integration.
Level 5Full ML Automation. Ability to come up with super-human strategies for solving hard ML problems without any input or guidance. Fully conversational interaction with the human user.

(credit: Bojan Tunguz)

The current state of AutoML is still too technical for a business user. This is where Ople comes in. The Ople Platform really makes the machine learning process simpler and faster, because many technical requirements are removed. Our proprietary AI analyzes and automates many more tasks other than machine learning model building. For example, the Ople Platform intelligently allocates resources to maximize efficiency and builds the final model that can be deployed and put into production within minutes after you upload or connect your dataset. As a user, you would simply prepare a dataset that describes your business and upload or connect the dataset to the platform. Once uploaded, The Ople Platform will ask a few questions for you to verify the objective of the challenge. There is no coding required at all. You would click a few times and the platform will intelligently find valuable insights for you. It will then automatically build a deployable model so you can start making predictions and perform What-If analysis on the platform, use API to connect the model into your business applications or Google Sheets or download the model.

“But how do I know your model is the best model?” The answer is simple. The world-renowned Kaggle grandmaster Giba is our Lead Data Scientist who built our AI engine. In a recent interview, Giba explained that the time he spent on Kaggle helped him to explore different machine learning algorithms and establish a good understanding of them, as well as more effective approaches to optimize hyperparameter tuning in different situations. All his expertise and intuition are embedded in the Ople Platform, so you as a user always have the best data scientist working with you. 

If you have strong doubts about this, we humbly ask you to contact us for a head-to-head challenge. If you are ready to make a difference in your business, request a free demo here.