At Seldon, we’re on a mission to deliver the most powerful enterprise machine learning platform, which means you solve real-world problems faster.
We’ve hit another milestone release with Seldon 0.97, which includes support for real-time feature transformations in our new predictive pipelines. We also share some secrets for advanced recommendation engine configuration. All of these new features are packaged for immediate use with virtual machines.
In the 0.97 release, Seldon extends your predictive pipeline with the ability to extract and transform the raw input features to allow you to build more effective models. Predictive pipelines have offline and real-time components as shown in the above diagram.
You can send raw data for creating a predictive pipeline via the REST API in real time. Data can be sent as arbitrary JSON, which gives you complete freedom to provide whatever data is available.
This raw JSON data is usually not in the best format to directly build machine learning models. Therefore, in the offline modelling stage the data is first sent through an (optional) set of feature transformations to extract and create appropriate features that are useful for creating predictive models. After these transformations, you can build a model to predict some target feature in the data based on the extracted/transformed features.
At runtime, you can now repeat the same set of offline transformations to create the same set of final features to test against the model. After the transformations, features can be scored and a predictive result is returned in real-time.
Our technical docs provide details of how to build feature extraction pipelines in Python. In future, we will also provide Spark-based pipelines to handle larger datasets. We also provide examples on creating models and runtime-prediction scorers with our Microservice API using two industry leading toolkits, Vowpal Wabbit and XGBoost.
Advanced Recommendation Configuration
- Seldon allows you to set up complex recommendation scenarios that are often needed in real production settings. Some examples include:
- To allow multiple algorithms to be tried in order until one succeeds for a user.
- To combine the results of multiple algorithms into a single recommendation result.
- To run A/B tests or multivariate tests to compare different algorithm strategies.
- To have multiple recommendations per page, e.g. in-section recommendations and site-wide recommendations.
- To serve different recommendations for mobile users as opposed to desktop users.
- To run a multivariate test on an API determined set of users, e.g. users who view a certain subsection of a web-site.
To find out more visit our online documentation.
Seldon packaged for immediate use
Seldon 0.97 open-source release is also available as a pre-built virtual machine, to get started quickly with implementations for a Vagrant VM and AWS AMI.
We love to hear your feedback and are happy to answer questions, so please don’t hesitate to send a message to the users group.