HTM In The Cloud (HITC)

Chat with us on Gitter:

About

This application provides a REST API for use with Nupic. Currently, it is meant to be deployed on Heroku.

Installation

Get a local instance working for testing purposes by cloning this repository and running run.

Once you've tested it out locally, you can deploy to heroku:

1. Install the heroku toolbelt
2. In the root directory of this project, run heroku create --stack cedar This will give the git repo a remote repo for heroku
3. Run:

   git push heroku master
   heroku scale web=1
   heroku ps # to see if it's running
   heroku logs -t

Your API should be ready to go!

API

All methods return a JSON response, if a method required a GUID (globally unique identifier), and there is no model associated with it, a 404 error will be returned.

While still in development the REST API is as follows:

/models

POST

Creates a model,

Parameters

Takes a dict of arguments in with the following fields:

• guid: an optional name for the model, by default a UUID4 will be generated. The guid must be unique.
• predictedField: the field to be predicted
• modelParams: the nupic model parameters to be used. If none are provided, default parameters are provided based on those from the hot gym anomaly detection.

Returns

A JSON object containing the GUID of the model and parameters used.

GET

Gets a list of all models and their associated information.

Returns

A JSON list of JSON objects containing:

• The GUID of the model
• The parameters used for the mdoel
• The predicted field
• The number of input records processed
• The last input record processed
• The temporal field if it exists

/models/{guid}

If the GUID is not found, a 404 error will be returned

GET

Returns

Details of the model (same format as GET /models)

DELETE

Deletes the model

Returns

404 if no such model exists. The body will be a JSON object with:

• success: true if model was deleted, false otherwise
• guid: the guid of the model to deleted

PUT

Run data through a model

Parameters

Takes a list of objects or a single object. If using a datetime field, then they must be pushed in chronological order. Each input object must have those fields given in the parameters. Timestamp must be an integer or float that is a unix timestamp.

Returns

Returns the output of the nupic model:

• anomaly score
• prediction
• likelihood

/models/reset/{guid}

Resets a model