peal.teachers.web_feedback_teacher¶
File-based feedback teacher for the web demo.
Human2ModelTeacher runs its own Flask server on a port and blocks until a
person has clicked through every collage. A job worker behind another web app
cannot do that, so this teacher talks to the web app through the job directory:
<feedback_dir>/feedback_request_<round>.json written by the teacher <feedback_dir>/feedback_response_<round>.json written by the web app
The request lists the collages that need a verdict (index, path, source and
target class, end confidence); the response maps index -> “true” | “false” |
“ood”. The teacher polls for the response file and returns the verdict list in
the order of collage_path_list, with the same sentinels Human2ModelTeacher
emits for pairs that never reach a person (“student incorrect!” when the
student was wrong on the original, “student not swapped!” when the edit did not
cross the boundary).
PEAL_WEB_FEEDBACK_AUTO = “true” | “false” | “ood” answers every request
without a person (tests, smoke runs).
Classes
Teacher that collects human verdicts through JSON files in a job directory. |