add on-task-bert as source for mathWord

New measures/OnTaskAnalyser.py (mirrors VocabularyAnalyser's shape:
self-contained, owns its own model instance, no shared handler
device/tokenizer/input-builder) runs edsi-umd/on-task-bert on student
utterances. moments.mathWord is now derived from that classifier
instead of bool(vocabulary_terms).

vocabularyTerms/vocabularyMatches/vocabularyTable are untouched -
VocabularyAnalyser keeps running as-is, since the frontend's
MathVocabularyTable needs the per-word match data the classifier
can't produce.

TODO: confirm the exact label string on-task-bert's pipeline returns
for "on-task" (see ON_TASK_LABEL in OnTaskAnalyser.py) with a real
inference call before relying on this in production.

handler.py

@@ -1,5 +1,6 @@
 from typing import Dict, List, Any
 from measures.VocabularyAnalyser import VocabularyAnalyser
+from measures.OnTaskAnalyser import OnTaskAnalyser, ON_TASK_LABEL
 from scipy.special import softmax
 from collections import Counter, defaultdict
 import numpy as np
@@ -47,6 +48,7 @@ class Utterance:
         self.num_math_terms = None
         self.math_terms = None
         self.vocabulary_terms = None
+        self.on_task = None
 
         # moments
         self.uptake = None
@@ -92,7 +94,7 @@ class Utterance:
                 'questioning': True if self.question else False, 
                 'uptake': True if self.uptake else False, 
                 'focusingQuestion': True if self.focusing_question else False,
-                'mathWord': bool(self.vocabulary_terms)
+                'mathWord': bool(self.on_task) and self.on_task.get('label') == ON_TASK_LABEL
             },
             'unitMeasure': self.unit_measure,
             'aggregateUnitMeasure': self.aggregate_unit_measure,
@@ -537,7 +539,11 @@ class EndpointHandler():
         vocabulary_analyser = VocabularyAnalyser(str(glossary_path))
         vocabulary_analyser.run_analysis(transcript)
         del vocabulary_analyser
-        
+
+        on_task_analyser = OnTaskAnalyser()
+        on_task_analyser.run_analysis(transcript, uptake_speaker=uptake_speaker)
+        del on_task_analyser
+
         transcript.update_utterance_roles(uptake_speaker)
         sorted_math_cloud, teacher_math_cloud, student_math_cloud = run_math_density(transcript)
         transcript.calculate_aggregate_word_count()

measures/OnTaskAnalyser.py

@@ -0,0 +1,23 @@
+from transformers import pipeline
+
+ON_TASK_MODEL = 'edsi-umd/on-task-bert'
+ON_TASK_LABEL = 'LABEL_1'  # TODO: confirm against a real inference call before relying on this
+
+
+class OnTaskAnalyser:
+    def __init__(self, model_path=ON_TASK_MODEL, max_length=256):
+        self.pipe = pipeline("text-classification", model=model_path,
+                              truncation=True, max_length=max_length)
+
+    def predict_one(self, text: str):
+        return self.pipe(text)[0]  # {'label': ..., 'score': ...}
+
+    def run_analysis(self, transcript, uptake_speaker=None):
+        """Mutate transcript utterances by setting on_task for student utterances."""
+        for utt in transcript.utterances:
+            if uptake_speaker is not None and utt.speaker == uptake_speaker:
+                continue  # model trained on student utterances only
+            if not utt.text or not utt.text.strip():
+                continue
+            utt.on_task = self.predict_one(utt.text)
+        return transcript