nb_actions returned by project_set_get_user_stats can decrease

[jiho]

In project 185 project_set_get_user_stats() returns (partial output)

[{'activities': [{'id': 106,
                 'last_annot': '2023-02-14T06:00:34',
                 'nb_actions': 500},
                {'id': 4,
                 'last_annot': '2023-03-22T14:45:06',
                 'nb_actions': 48495},
                {'id': 993,
                 'last_annot': '2023-03-22T14:48:46',
                 'nb_actions': 5},
...
 'annotators': [{'id': 106, 'name': 'Annotateur'},
                {'id': 4, 'name': 'Jean-Olivier Irisson'},
                {'id': 993, 'name': 'Test API user'},

I have ~48,500 actions while the project has 8000 objects. I suspect that many of these actions are due to numerous "reset to predicted".

When I perform new classifications, nb_actions does not change (note the change of last_annot but no change in nb_actions

[{'activities': [{'id': 106,
                 'last_annot': '2023-02-14T06:00:34',
                 'nb_actions': 500},
                {'id': 4,
                 'last_annot': '2023-03-22T14:55:15',
                 'nb_actions': 48495},
                {'id': 993,
                 'last_annot': '2023-03-22T14:48:46',
                 'nb_actions': 5},

When user 993 performs classifications, his nb_actions increases, mine decreases (by 2 here):

[{'activities': [{'id': 106,
                 'last_annot': '2023-02-14T06:00:34',
                 'nb_actions': 500},
                {'id': 4,
                 'last_annot': '2023-03-22T14:55:15',
                 'nb_actions': 48493},
                {'id': 993,
                 'last_annot': '2023-03-22T14:56:52',
                 'nb_actions': 7},

I would expect nb_actions to never decrease.

[grololo06]

It is most likely due to current state of the objects (which counts as 1 action) and the fact that we don't always keep history, so new classif (overriding current state) + no history kept -> decreasing count.
I would need to know which classifications were made, manual, prediction?

[jiho]

Not sure I get this:

It is most likely due to current state of the objects (which counts as 1 action) and the fact that we don't always keep history, so new classif (overriding current state) + no history kept -> decreasing count.

completely.

The new classif in the tests above were manual classifications, as me or as Test API user