New capabilities in Release 1.10 further accelerates the deployment process by streamlining model on-boarding into FastScore. Through FastScore’s powerful but lightweight abstractions, data scientists can encapsulate their models for easy deployment through the FastScore System without any re-coding or refactoring. Release 1.10 streamlines conformance to FastScore’s abstractions for a large number of additional classes of models. Additionally, Release 1.10 adds automated schema inference to allow data scientists to quickly generate the externalized schemas to ensure compatibility of data streams with the model.
Additional “model conformance” approaches to support a larger variety of data scientist workflows and programming paradigms. For particular models, this approach streamlines the process for adding a new model into FastScore by allowing data scientists to simply change their model’s input and output data structures to read from the FastScore engine “slots.” A typical paradigm that data scientists use to structure their code includes: (1) reading in data (2) processing and scoring the data (3) writing the inference out to a consuming application or data feed. Through this feature, data scientists need only make minor modifications to steps 1 and 3, using the intuitive FastScore libraries. Note that FastScore will continue to support the prior “conformance” approaches, which are still well suited for certain classes of models (e.g. explicit RESTful models). This capability includes support for:
Added a schema inference utility to generate the FastScore-compliant data schema automatically from a sample data file.
Added full support for a Go SDK for core FastScore interactions, allowing data scientists, data engineers, model ops engineers, and others to leverage the rapidly-growing GO language for standard interactions and management of FastScore. We have also used the new Go SDK to improve our CLI so it no longer depends on python implementations, allowing for more portability.
Enhanced the core logging framework to leverage the ELK (Elasticsearch, Logstash, Kibana) stack, providing system architects and support engineers more flexibility to integrate FastScore’s logging and monitoring capabilities into existing enterprise systems and standards.