proxyml 0.7.0__tar.gz → 0.9.0__tar.gz

{proxyml-0.7.0 → proxyml-0.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: proxyml
-Version: 0.7.0
+Version: 0.9.0
 Summary: Python SDK for calling the ProxyML API
 Author-email: ProxyML <contact@proxyml.ai>
 License:                                  Apache License
@@ -211,7 +211,7 @@ Requires-Dist: pandas>=1.5
 Requires-Dist: orjson>=3.9
 Requires-Dist: proxyml-core>=0.1
 Provides-Extra: local
-Requires-Dist: proxyml-core[modeling]>=0.2.1; extra == "local"
+Requires-Dist: proxyml-core[modeling]>=0.2.2; extra == "local"
 Provides-Extra: dev
 Requires-Dist: build; extra == "dev"
 Requires-Dist: twine; extra == "dev"

{proxyml-0.7.0 → proxyml-0.9.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "proxyml"
-version = "0.7.0"
+version = "0.9.0"
 description = "Python SDK for calling the ProxyML API"
 readme = "README.md"
 license = {file = "LICENSE"}
@@ -35,7 +35,7 @@ dependencies = [
 
 [project.optional-dependencies]
 local = [
-    "proxyml-core[modeling]>=0.2.1",
+    "proxyml-core[modeling]>=0.2.2",
 ]
 dev = [
     "build",

{proxyml-0.7.0 → proxyml-0.9.0}/src/proxyml/local/__init__.py

@@ -10,6 +10,7 @@ from proxyml.local.challenger import (
     Complexity,
     Rung,
     TrainedChallenger,
+    fingerprint_labels,
     score_champion,
     to_challenger_upload,
     train_auto_challenger,
@@ -21,6 +22,7 @@ __all__ = [
     "train_auto_challenger",
     "score_champion",
     "to_challenger_upload",
+    "fingerprint_labels",
     "Complexity",
     "Rung",
     "TrainedChallenger",

{proxyml-0.7.0 → proxyml-0.9.0}/src/proxyml/local/challenger.py

@@ -27,6 +27,7 @@ from sklearn.pipeline import Pipeline
 
 from proxyml.schema_builder import get_schema
 from proxyml_core.export import SurrogateExport
+from proxyml_core.fingerprint import FINGERPRINT_MISMATCH_MESSAGE, check_fingerprints
 from proxyml_core.modeling.estimators import (
     binarize_if_probabilities,
     extract_hyperparameters,
@@ -127,7 +128,7 @@ class TrainedChallenger:
     champion_metrics: dict[str, float] | None = None
 
 
-def _fingerprint_values(values: np.ndarray | list) -> str:
+def fingerprint_labels(values: np.ndarray | list) -> str:
     """Hash an array of labels, deterministically, without the data ever leaving this process.
 
     ``.tolist()`` converts numpy scalars to native Python types before
@@ -135,6 +136,13 @@ def _fingerprint_values(values: np.ndarray | list) -> str:
     different scalar repr behavior. Order is preserved (not sorted) since
     it encodes row alignment — that's exactly what a "same data?" check
     needs to be sensitive to.
+
+    Public so callers that score a champion decoupled from
+    ``train_challenger()`` — e.g. two separate tool calls in an MCP
+    server, with no shared Python state between them — can compute
+    ``champion_data_fingerprint`` themselves and carry it into a payload
+    assembled by hand, the same way ``to_challenger_upload(champion_labels=...)``
+    does internally.
     """
     canonical = json.dumps(np.asarray(values).tolist())
     return hashlib.sha256(canonical.encode("utf-8")).hexdigest()
@@ -202,7 +210,7 @@ def train_challenger(
             and the result is attached as ``TrainedChallenger.champion_metrics``.
     """
     target_arr = np.asarray(target)
-    target_fingerprint = _fingerprint_values(target_arr)
+    target_fingerprint = fingerprint_labels(target_arr)
     if champion_predictions is not None and len(champion_predictions) != len(target_arr):
         raise ValueError(
             f"champion_predictions must have one entry per row of target "
@@ -343,6 +351,13 @@ def to_challenger_upload(
     itself still requires ``champion_metrics`` at upload time; this function
     just doesn't force you to have it up front.
 
+    If both fingerprints end up present and don't match, raises
+    ``ValueError`` immediately — via the same
+    ``proxyml_core.check_fingerprints``/``FINGERPRINT_MISMATCH_MESSAGE`` the
+    upload endpoint itself uses — rather than waiting for a round-trip to
+    the server to find out the challenger and champion were scored on
+    different data.
+
     Args:
         result: output of ``train_challenger()``/``train_auto_challenger()``.
         n_samples: size of the evaluation set both ``result.metrics`` and
@@ -390,9 +405,13 @@ def to_challenger_upload(
         payload["champion_metrics"] = champion_metrics
         payload["challenger_data_fingerprint"] = result.target_fingerprint
         if champion_labels is not None:
-            payload["champion_data_fingerprint"] = _fingerprint_values(champion_labels)
+            payload["champion_data_fingerprint"] = fingerprint_labels(champion_labels)
         elif used_internal_champion_metrics:
             payload["champion_data_fingerprint"] = result.target_fingerprint
+        if check_fingerprints(
+            payload["challenger_data_fingerprint"], payload.get("champion_data_fingerprint")
+        ) is False:
+            raise ValueError(FINGERPRINT_MISMATCH_MESSAGE)
     return payload
 
 

{proxyml-0.7.0 → proxyml-0.9.0}/src/proxyml.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: proxyml
-Version: 0.7.0
+Version: 0.9.0
 Summary: Python SDK for calling the ProxyML API
 Author-email: ProxyML <contact@proxyml.ai>
 License:                                  Apache License
@@ -211,7 +211,7 @@ Requires-Dist: pandas>=1.5
 Requires-Dist: orjson>=3.9
 Requires-Dist: proxyml-core>=0.1
 Provides-Extra: local
-Requires-Dist: proxyml-core[modeling]>=0.2.1; extra == "local"
+Requires-Dist: proxyml-core[modeling]>=0.2.2; extra == "local"
 Provides-Extra: dev
 Requires-Dist: build; extra == "dev"
 Requires-Dist: twine; extra == "dev"

{proxyml-0.7.0 → proxyml-0.9.0}/src/proxyml.egg-info/requires.txt

@@ -12,4 +12,4 @@ black
 ruff
 
 [local]
-proxyml-core[modeling]>=0.2.1
+proxyml-core[modeling]>=0.2.2

{proxyml-0.7.0 → proxyml-0.9.0}/tests/test_local_challenger.py

@@ -8,6 +8,7 @@ from proxyml.local import (
     Complexity,
     LADDERS,
     TrainedChallenger,
+    fingerprint_labels,
     score_champion,
     to_challenger_upload,
     train_auto_challenger,
@@ -446,18 +447,15 @@ def test_to_challenger_upload_champion_labels_fingerprint_for_decoupled_path():
     assert payload["champion_data_fingerprint"] == result.target_fingerprint
 
 
-def test_to_challenger_upload_champion_labels_fingerprint_differs_for_different_data():
+def test_to_challenger_upload_raises_on_fingerprint_mismatch():
     df = _labeled_df(seed=33)
     target = df["approved"]
     result = train_challenger(df, target, _schema(), task="classification")
     champion_metrics = score_champion(target, target, task="classification")
     other_labels = ~target
 
-    payload = to_challenger_upload(
-        result, champion_metrics=champion_metrics, champion_labels=other_labels
-    )
-
-    assert payload["challenger_data_fingerprint"] != payload["champion_data_fingerprint"]
+    with pytest.raises(ValueError, match="don't match"):
+        to_challenger_upload(result, champion_metrics=champion_metrics, champion_labels=other_labels)
 
 
 def test_to_challenger_upload_omits_champion_fingerprint_without_labels_or_internal_path():
@@ -480,3 +478,14 @@ def test_to_challenger_upload_without_champion_metrics_omits_fingerprints():
 
     assert "challenger_data_fingerprint" not in payload
     assert "champion_data_fingerprint" not in payload
+
+
+def test_fingerprint_labels_is_public_and_matches_target_fingerprint():
+    """Exercises the exact use case this was made public for: a caller with
+    no TrainedChallenger in hand (e.g. a decoupled score_champion() call)
+    computing the same fingerprint train_challenger() would have."""
+    df = _labeled_df(seed=36)
+    target = df["approved"]
+    result = train_challenger(df, target, _schema(), task="classification")
+
+    assert fingerprint_labels(target) == result.target_fingerprint