Distributed analytics platform technology that is organically distributed with your data

What's in the Box?

Node

Intelligent software agent co-located with a data source. Nodes provide a web API for deploying locally learned predictive models to the Node. Once deployed, these local models are discoverable within the platform by all Nodes. As autonomous agents, Nodes:

• Collaborates with other Nodes to learn prediction network models
• Queries its scoring engine to generate a local score and collaborates with other Nodes using the prediction network model to generate the network prediction

Each Node serves a Collaborative Analytics web application providing visual interfaces to:

• Manage Prediction Networks: discover related business queries and assemble prediction networks by selecting the models that answer these queries at different data sources as participants.
• Prediction Network Query: submit queries to the prediction network and visualize the prediction results.

ML Toolbox Plugin

Machine learning language bindings that implement the local model deployment protocol to the Node and provide a user interface for such deployment. Current bindings include:

• R: prediction models built using rattle and caret packages
• Python: models built using the scikit-learn package (Coming Soon)

The Collaborative Analytics platform leverages your existing predictive analytics toolset and workflow, providing both ease and low cost of adoption

There is no new language to learn, no new machine learning package to master, and no steep learning curve to climb. Collaborative Analytics augments the familiar Predictive Analytics Lifecycle, allowing your organization to continue using the same analytics tools and workflow, in which they are invested, to build predictive models answering your business queries of the data. The additional steps in Collaborative Analytics let your data analysts rapidly discover related models answering related business queries across the network to assemble and to evaluate ever larger prediction networks of these models with value-added results towards your business goals.

Deploy Local Model

In Collaborative Analytics, the deployment step includes deploying the evaluated performance of the local model to the Node, which is used subsequently by the Node in collaboration with other Nodes to learn prediction networks for which the model is a participant. The ML Toolbox Plugin provides a visual interface for this deployment. In submitting the model, the data scientist attaches metadata in the form of keywords and description to allow the business query answered by the model to be discoverable across the network via distributed search.

Discover Related QuerieS

Once the model is deployed to the Node, the business query that it answers is discoverable across the network. The data analyst uses the Collaborative Analytics web application to search for related queries and the models that answer them. The Collaborative Analytics web application provides a visual interface to examine a returned related query and its model properties, including the descriptions of the related business query, the data source, and the performance of the model in answering it.

Assemble Prediction Network

Within the Collaborative Analytics web application, any subset of related business queries, that includes one's own, can be selected as participants of a prediction network to augment the answering of one's own query. New prediction networks are created in this manner. Once participants are selected, the prediction network is learned automatically by the collection of participating Nodes, resulting in a projected accuracy for the network upon completion. This automated process is orders of magnitudes faster than learning the model from centralized data. The visual interface makes it easy to add or delete participants in a network, making quick exploration of different combinations of data sources possible across cycles of the Collaborative Analytics process.

Monitor Performance

Within the same Collaborative Analytics web application, prediction networks created across the platform are queried by the data analyst to generate predictions. Prediction results from the network are then collected by the web application and visually displayed to the analyst. For test datasets with labeled truths, the Prediction Network Query web application computes the resulting performance over the dataset. This supports rapid testing and comparison of different prediction networks answering the same business query with different combinations of participating data sources, all using the same test dataset. 

To request a remote or live demonstration and discussion, please fill out our Contact Request Form.