chore: establish project baseline

scripts/bulk_classify_in_container.py

@@ -0,0 +1,46 @@
+"""Runs INSIDE the fiscal-hub container (read-only on PostgreSQL): classifies [ncm, description] pairs with the
+production Classifier through a memoizing repository wrapper. Same code path as /v1/classify, minus HTTP and the
+classification_cache write; the wrapper only memoizes pure dataset lookups (datasets are static during the run).
+usage: python /tmp/bulk_classify_in_container.py /tmp/in.json /tmp/out.json
+"""
+import json
+import sys
+from functools import lru_cache
+
+from central.app.classification import Classifier
+from central.app.datasets.service import DatasetService
+from central.app.db import SessionLocal
+
+
+class Memo:
+    def __init__(self, service):
+        self._s = service; self._cache = {}
+
+    def __getattr__(self, name):
+        attr = getattr(self._s, name)
+        if not callable(attr):
+            return attr
+        def wrapped(*a, **kw):
+            key = (name, a, tuple(sorted(kw.items())))
+            try:
+                hash(key)
+            except TypeError:
+                return attr(*a, **kw)
+            if key not in self._cache:
+                self._cache[key] = attr(*a, **kw)
+            return self._cache[key]
+        return wrapped
+
+
+def main():
+    pairs = json.load(open(sys.argv[1], encoding="utf-8"))
+    out = []
+    with SessionLocal() as session:
+        classifier = Classifier(Memo(DatasetService(session)))
+        for ncm, desc in pairs:
+            out.append(classifier.classify({"current_ncm": ncm, "description": desc}))
+    json.dump(out, open(sys.argv[2], "w", encoding="utf-8"), ensure_ascii=False, default=str)
+    print(len(out))
+
+
+main()