Define a classifier XMLAnnotation format

[spli]

Define an XMLAnnotation format, as used for example with the multi-dimensional Modulo work
http://www.openmicroscopy.org/site/support/file-formats/working-with-ome-xml/6d-7d-and-8d-storage

This will probably have two main requirements.

Storage of trained classifier instances

Store classifier parameters and trained state:
For example, in
​https://github.com/manics/omero-pychrm/tree/20130228-pi_dev_demo
PyCHRM classifiers are stored in the form of three tables which are attached to the project containing the training data:

• Feature weights
• Class labels/indexes
• Selected features values (nearest neighbour means the values of the selected original features are required, along with the original class label).

In addition we should probably store the classifier parameters, and possibly some sort of versioning information.

Storage of classifier results

Store classifier predictions, possibly multiple per image/object. In ​https://github.com/manics/omero-pychrm/tree/20130228-pi_dev_demo the idea of tagsets containing the labels associated with a classifier instance was used (#10367).

We could use an XMLAnnotation to store the final predictions from running a classifier instance, for instance by listing image IDs instead of tagging them. This gives us more flexibility than using tagsets which are limited to a single level, and which are defined by a namespace so it's not possible to also have a classifier namespace. It is also extensible to classifiers which have parameters controlling the prediction as well as the training, since these additional parameters can be easily recorded.

[spli]

Initial draft for the classifier XMLAnnotation here:
​https://gist.github.com/manics/5238573

[jamoore]

a5d34ad53b1b23de9c629887769c1dc0305f4746 (second revision) certainly looks very easy to read. Will be interesting to see how difficult this will be for you to work with.

[jburel]

Moved from sprint 2013-04-09 (7))

[jburel]

Moved from sprint 2013-05-07 (8)