AIDR Release 1.1.0
Main New features:
- Re-design of AIDR homepage with tiled buttons, multiple UI modifications.
Administrator can make a collection publicly viewable or private only. - Users can now delete a collection.
- A new publicly browsable collection page created that allows users to browse the available collections and download data without the need to login.
- Crisis creation now imposes unique constraint on collection names.
- Crisis type is now a mandatory field for creating a crisis.
- Rule for replacement of a classifier by a new model has been changed.
- Rule for Pybossa scheduler has been updated - PyBossa Scheduler now never leaves the document table "empty".
- Live data feed on AIDR main page uses a balanced sampling to prevent starvation of tweets from minority feeds and filters low-confidence tweets.
- Configurable skin for crowdsourcing tasks.
- Upgraded Pybossa, now has a Mail server to serve forgot pwd or reset password functions.
- Added new default classifiers.
- The task_buffer view in the database in now deprecated.
- New algorithm for truncating documents from the aidr_predict.document table by the Tagger that avoids locking the database for long durations.
- The document and task_assignment tables in the aidr_predict database are now periodically cleaned to remove documents that are too old (duration is user configurable).
- New module aidr-task-manager added to provide database abstraction - all calls that touch any of the following aidr_predict database tables will in the future pass only through this module:
- document
- task_assignment
- document_nominal_label
- Tagger-API now now uses aidr-task-manager to access the relevant tables of the aidr_predict database.
- New interactive view/download page available for filtering classified tweets based on user-defined criteria.
- Persister module upgraded with new CSV download APIs to support filtering of tweets based on user-defined criteria.
- Output module upgraded with new output APIs to support filtering of tweets based on user-defined criteria.
Bug Fixes from release 1.0.0:
- Problem with CSV download due to AJAX time-outs.
- Persister crash issues.
- Classifier creation defect that would create partial classifier when tagger name is too long.
- Users should not be able to edit in the language combo-box in edit/create collection page.
- Trainer UI would show NaN more needed to re-train.
- Misreporting in number of already-classified tweets.
- Remove near-duplicate CSS files from css folder.
- Page with classifier details does not render correctly if you are not a sysadmin.
- Tagger error on displaying and/or save.
- Deadlock issue in aidr_predict database.
- Deleting a classifier should notify pybossa-scheduler.