The Protocols and Structures for Inference project aims to develop a general purpose web API for machine learning. This JavaScript client provides an 'explorer' interface for communicating with full-blown PSI services and individual PSI resources (relations, attributes, learners, predictors, etc.).
It has been designed to support demonstrations of the API's behaviour to technically-minded, often machine learning-oriented, people. It is thus not a production-quality end user application, but parts of the system could be used to build one.
The app uses RequireJS for module management. In addition to the widely-used jQuery, Backbone and Underscore packages, it uses the JSON Form package to generate and validate forms using JSON Schema and a version of Geraint Luff's Tiny Validator (for v4 JSON Schema). The tv4 package has been modified to report multiple errors, instead of just the first one, so that it can be used in place of JSV, which only supports JSON Schema up to draft version 3.
The client includes a comparison tool that illustrates how one or more predictors may be evaluated via a series of interactions with the server. This tool is relatively separate from the 'explorer' part of the client.
To demonstrate what the internet might look like with ubiquitous machine learning services, in compliant browsers the app requests to register a protocol handler for web+psi links. It uses BNC Connector to communicate between a handler page and an open client app, since the browser will open a new page to process each link, whereas the client has been designed to work with multiple clients simultaneously.
Note that, due to security restrictions, the protocol handler (specified in js/web+psi.js) must reside at the same location from which the client is being served.
Assuming you have previously registered a protocol handler for web+psi links, the following should load the project's demonstration service into a new client or an already open client:
web+psi://poseidon.cecs.anu.edu.au
Further documentation may be provided in the future.