Add Docker configuration

3.4

documentation changes

Documentation updates

Minor changes in Swagger documentation regarding data requirements for component usage

Doc changes + bug fixing

Includes bug fixing from version 3.3.1 + doc updates

Bug fixing

Solves bugs from previous version release.

Extend multiclass logic

This version does not include any interface changes.

The requirements classifier v3.3 has been extended and fixed to allow multiclass models to all of its use cases, including the train&test, which were limited to nº of classes = 2 before.

For previous integrations this version does not affect to its performance.

Fix delete interface

Parameters 'company' and 'property' from DELETE /model methods have been refactored from the request body to query parameters due to interface design consistency.

New confidence computation

This release includes a new algorithm to evaluate the confidence value of the /classify method.

The new confidence value is calculated based on the global score given by the Mahout classifier tool for each tag of the classifier (positive and negative).

Given a requirement R and a tag T with two possible classification values (T and ^T), the service computes the difference between T and ^T scores of the classifier, and applies a normalization based on statistical analysis. For each property value recommendation returned by the classifier, a confidence X is returned, which is a value between 50% and 100% - as the returned recommendation is always the most probable property.

Contextual analysis + Async communication

Version 3.0 of requirements-classifier includes the following new features:

• Train classifier methods (both binary and multilabel) are now asynchronous using the observer pattern. After sending the request the client will get a response containing an ID which identifies the request, which will be processed on another thread. Once the process finishes, the response will be sent to an endpoint set by the client. Details about this feature can be checked on swagger documentation. For users of v2.0 notice that this will break their integrations.

• Contextual information analysis implemented. No interface changes regarding v2.0, but new version takes into account contextual information by applying a generic unique role for requirement items belonging to a same list.

• List of model values created in multilabel classifier. There is a new endpoint which allows clients to get a list of the models created given a property with multilabel classification.

• Removed creation of empty models. If no data is sent for a specific label, the multilabel classifier will no longer create a model for this property. This is to avoid confussion as to the purpose of a classifier, so an automatic rejection classifier is not created.

Multi-domain classifier + API Refactor

2.0

fixed multi domain recommendations

Generic multi-classifier implementation

The requirements-classifier service exposes two controllers:

• binary-classification-controller: handles binary classifications as in previous versions, given a specific company and property to classify on the set of requirements. No changes have been applied to binary classification

• multi-classification-controller: this is a refactor of the previous domain-based-controller. We have modified both controller and logic of the multidimensional classifier so that it is generic to any UC and property desired to use for this purpose. Main changes are:

  1. reqDomains property has been removed; the component expects now, for each requirement, a custom key referred as property-key, which is used for the multidimensional classification. The tag of this key is sent as a parameter to the petitions. So, for instance, in orther to use the component as in previous versions, it is necessary to send as a query param property-key=reqDomains, and no changes are required on the JSON object.

  2. classify method of the multi-classification-controller is now a fully-multidimensional classification process. Instead of classifying the requirements against a single value of the property-key, now all models of the given property are used to classify the requirements. For each requirement, a list of the matching assignations is returned, as the level of confidence which is the mean of all the recommendations.

Please reffer to https://api.openreq.eu/requirements-classifier/swagger-ui.html#/ to check details about the new API methods.