Review: H2O.ai automates machine learning

Driverless AI really is able to create and train good machine learning models without requiring machine learning expertise from users

1 2 Page 2
Page 2 of 2

Evaluating automated machine learning

Overall, Driverless AI is impressive—in fact, I’m surprised that it works so well. The company says that its own Kaggle masters, who provided the algorithms for the system, were also surprised. Feature engineering and model training often takes weeks before you get a good answer. Driverless AI can often get a good answer in minutes or hours.

H2O.ai claims that Driverless AI brings you the intelligence of a Kaggle grandmaster in a box. Well, sort of. It brings you the methods of a Kaggle grandmaster, but somebody in your organization does need to understand something about what you’re doing or you won’t be able to make the decisions that Driverless AI leaves to you, such as what columns to drop, what accuracy and time to choose, and what scoring algorithm to choose.

Without some statistical background, the discussions of feature engineering, k-means clusters, and generalized linear models will all sound like word salad, and the visualizations won’t spur you to say things like “Hmm, those two variables are highly correlated—I wonder if the model would come out better if I drop one.”

As far as I know, Driverless AI is the only shipping, supported, automatically driven machine learning system that also does feature engineering and annotation at the present time. I don’t think that will be the case for more than a year, though. Competition is cooking, and some of it might be released as open source or offered as a service.

Meanwhile, at $75K per GPU per year, Driverless AI is not cheap. It’s probably worth the price, however, if it is used by multiple data analysts and your company has many classification or regression problems that affect your bottom line in six-figure or seven-figure ways.

Cost: $75K per GPU per year, with discounts for four and eight GPU configurations. Equivalent CPU pricing for those who don’t have GPUs.

Platform: Ubuntu, RHEL, MacOS, Windows, or IBM Power; on-prem or any cloud; Chrome browser. Recent Nvidia GPUs (K80 or later) supported on Ubuntu 16.04.

At a Glance
  • H2O.ai's Driverless AI is an automatically driven machine learning system that also does feature engineering and annotation, dramatically reducing the time and effort required to produce good models.

    Pros

    • Driverless AI is able to create and train good models without requiring user expertise
    • Good integration with Nvidia GPUs (K80 and above)
    • Approximate linear models help to explain important factors in a decision
    • Makes quick work of generating and evaluating many models
    • Generates and exports prediction pipeline for trained model

    Cons

    • While the H2O.ai AI platform is open source, Driverless AI is proprietary
    • The concepts behind Driverless AI require a strong statistics and machine learning background
    • Trained data scientists will most likely be able to do more with Driverless AI than business analysts

Copyright © 2017 IDG Communications, Inc.

1 2 Page 2
Page 2 of 2