“It’s important for any independent AI platform to run effortlessly on the Google Cloud Platform, Microsoft Azure, and AWS,” said Forrester analyst Mike Gualtieri. “Having DataRobot in the Google Cloud Marketplace will allow customers to try and then use the platform and scale it easily to GCP.”
DataRobot is on Azure, AWS and GCP
The AI Cloud Platform is now available on Azure, while the DataRobot Automated Machine Learning platform is accessible on AWS.
Announced on June 7, 2022, the full availability of its solutions in the marketplace of the third largest cloud service provider is a milestone that DataRobot needs to reach. And this even though the publisher is competing with GCP’s own AI services, according to Mike Gualtieri.
“From a market point of view, I don’t think it affects DataRobot’s competitiveness,” the analyst observes. “It’s something they have to do. They are not ahead. They are not late. “
AWS, Microsoft and Google are eager to accept smaller competitors as it expands the ecosystems of these tech giants, he politely asserted.
More precisely, these cloud specialists are basing their economic growth on data storage and the implementation of increasingly complex processing. In principle for these actors, it doesn’t matter if the software used is their own, as long as the company deploying it relies on their computing and storage opportunities.
For the publisher, this is a great way to reach customers who frequently use a data warehouse or data lake from these suppliers. For investors, it’s the promise of seeing the DataRobot coffers refilled monthly through monthly subscriptions.
It remains to be seen if the pricing conducted by the publisher will convince customers regarding the features offered. Subscriptions to the DataRobot Platform in Google Cloud, which include AutoML and MLOps functionality, start at $ 5,416.67 per month or $ 65,000 per year.
IA Cloud: DataRobot prepares for production deployments
In addition to GitHub, DataRobot said it has integrated IT monitoring and security solutions from Sumo Logic, Splunk and Datadog, as well as customer service tools from Zendesk.
Although DataRobot has introduced many updates and new capabilities for the AI Cloud, these enhancements are widely expected, Mike Gualtieri reports.
And for good reason, the publisher changed its update cycle by moving from a quarterly release rate to another, monthly. Just two months have passed between the availability of DataRobot AI Cloud 8 and the May 2022 AI Cloud release.
New features include what the publisher calls “code-first” notebooks integrated into the platform, which allow data scientists to write their models in Python, Spark and R. The associated interface also allows import models from Jupyter or Apache Zeppelin.
This technology, currently in preview, came from the acquisition of DataRobot, in May 2021, by Zepl, the publisher of an analytical platform.
Additionally, DataRobot has added geospatial capabilities to its no-code predictive AI applications. The publisher also highlighted some extended capabilities. These include:
- Automated compliance documentation for all AI and machine learning models, including those developed outside of DataRobot;
- bias mitigation capabilities that automatically define and adapt models before deployment;
- MLOps features to support full model lifecycles and provide open models that IT teams can connect to third-party software platforms and services;
- better support for DevOps skills, with integration with GitHub Actions to automate machine learning workflows consistent with the CI/CD approach.
In March 2022, analysts wondered whether the direction taken by DataRobot would be satisfactory for data scientists reluctant to manipulate tools without code. Finally, the editor seems to be looking for a balance between the two aspirations.
Thus, Composable ML provides visual access to the hyperparameters that drive the natural language processing model. At the same time, enhanced automatic language discovery on Autopilot should accelerate the development of NLP solutions. Again, data scientists can observe the hyperparameters involved in the model classification decision (yes, NLP is usually just a series of classification tasks).
In addition, a visual tool allows you to filter the results of computer vision models. This allows to quickly understand why above or below a certain threshold the classification of an image passes or fails.
In the AI Cloud, DataRobot offers a function called XEMP, another interpretation method accessible from a CLI to try to understand the results of multiclass prediction algorithms. XEMP is an alternative to SHAP, a library focused on AI interpretability, which gave birth to the open source project SHAPASH on MAIF (approximately 1,800 stars on GitHub).
A bad pass for DataRobot?
Beyond this problem between offer completeness and ease of use, DataRobot should not only compete with cloudist solutions, but also with other independent publishers including H2O.ai, Dataiku, Databricks or SAS.
On paper, all the cards are in DataRobot to meet the challenge. It has raised $ 1 billion since inception, acquired seven smaller publishers, and reached a market valuation of $ 6.3 billion. However, the engine of success seems to be stuck. DataRobot has experienced significant setbacks in the past two.
In March 2020, at the start of the COVID-19 pandemic, DataRobot reduced its workforce without disclosing the number of employees affected by this layoff.
Recently, in early May, the publisher laid off about 70 employees, or 7% of its workforce, amid a market slowdown affecting most tech companies.