proxyml 0.3.0__tar.gz → 0.4.0__tar.gz

{proxyml-0.3.0 → proxyml-0.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: proxyml
-Version: 0.3.0
+Version: 0.4.0
 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]>=0.1; extra == "local"
+Requires-Dist: proxyml-core[modeling]>=0.2; extra == "local"
 Provides-Extra: dev
 Requires-Dist: build; extra == "dev"
 Requires-Dist: twine; extra == "dev"

{proxyml-0.3.0 → proxyml-0.4.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "proxyml"
-version = "0.3.0"
+version = "0.4.0"
 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]>=0.1",
+    "proxyml-core[modeling]>=0.2",
 ]
 dev = [
     "build",

{proxyml-0.3.0 → proxyml-0.4.0}/src/proxyml/__init__.py

@@ -1,3 +1,5 @@
+from importlib.metadata import version as _pkg_version
+
 from proxyml.client import (
     health_check,
     put_schema,
@@ -27,7 +29,10 @@ from proxyml.client import (
 )
 from proxyml.schema_builder import get_schema
 
+__version__ = _pkg_version("proxyml")
+
 __all__ = [
+    "__version__",
     "health_check",
     "put_schema",
     "fetch_schema",

{proxyml-0.3.0 → proxyml-0.4.0}/src/proxyml/local/__init__.py

@@ -10,6 +10,7 @@ from proxyml.local.challenger import (
     Complexity,
     Rung,
     TrainedChallenger,
+    score_champion,
     train_auto_challenger,
     train_challenger,
 )
@@ -17,6 +18,7 @@ from proxyml.local.challenger import (
 __all__ = [
     "train_challenger",
     "train_auto_challenger",
+    "score_champion",
     "Complexity",
     "Rung",
     "TrainedChallenger",

{proxyml-0.3.0 → proxyml-0.4.0}/src/proxyml/local/challenger.py

@@ -10,7 +10,7 @@ training target was real ground truth or a black box's predictions.
 
 from __future__ import annotations
 
-from dataclasses import dataclass
+from dataclasses import dataclass, replace
 from enum import Enum
 from pathlib import Path
 from typing import Any, Callable, Literal
@@ -19,7 +19,6 @@ import numpy as np
 import pandas as pd
 from sklearn.base import BaseEstimator
 from sklearn.linear_model import LogisticRegressionCV, RidgeCV
-from sklearn.metrics import f1_score, r2_score
 from sklearn.model_selection import train_test_split
 from sklearn.pipeline import Pipeline
 
@@ -34,6 +33,7 @@ from proxyml_core.modeling.estimators import (
 )
 from proxyml_core.modeling.extract import extract_export_data
 from proxyml_core.modeling.preprocess import build_preprocessor
+from proxyml_core.modeling.scoring import score_predictions
 from proxyml_core.schema import Feature, FeatureSchema
 
 
@@ -184,14 +184,10 @@ def train_challenger(
     hyperparameters = extract_hyperparameters(pipeline.named_steps["estimator"])
     y_pred = pipeline.predict(X_test)
 
-    metrics: dict[str, float] = {}
-    if classification:
-        metrics["f1"] = float(f1_score(y_test, y_pred, average="weighted", zero_division=0))
-        metrics["accuracy"] = float((y_test == y_pred).mean())
-    else:
-        metrics["r2"] = float(r2_score(y_test, y_pred))
+    metrics = score_predictions(y_test, y_pred, task=resolved_task)
 
     export = extract_export_data(pipeline, features, resolved_task)
+    export = replace(export, hyperparameters=hyperparameters, metrics=metrics)
 
     return TrainedChallenger(
         pipeline=pipeline,
@@ -202,6 +198,28 @@ def train_challenger(
     )
 
 
+def score_champion(
+    labels: np.ndarray | list,
+    predictions: np.ndarray | list,
+    *,
+    task: Literal["classification", "regression"],
+) -> dict[str, float]:
+    """Score a champion model's predictions against real labels, locally.
+
+    Uses the exact same scoring code as ``train_challenger()``'s internal
+    fidelity metrics, so ``champion_metrics`` and a paired
+    ``TrainedChallenger.metrics`` are computed identically — required for an
+    apples-to-apples champion-vs-challenger comparison. Pass the same task
+    the paired challenger resolved to (e.g. ``result.task``); there's no
+    ``task="auto"`` here, since letting the two resolve independently risks
+    them silently diverging.
+
+    Returns ``{"f1":..., "accuracy":...}`` for classification or ``{"r2":...}``
+    for regression — the same shape as ``TrainedChallenger.metrics``.
+    """
+    return score_predictions(np.asarray(labels), np.asarray(predictions), task=task)
+
+
 def train_auto_challenger(
     data: str | Path | pd.DataFrame,
     target_col: str,

{proxyml-0.3.0 → proxyml-0.4.0}/src/proxyml.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: proxyml
-Version: 0.3.0
+Version: 0.4.0
 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]>=0.1; extra == "local"
+Requires-Dist: proxyml-core[modeling]>=0.2; extra == "local"
 Provides-Extra: dev
 Requires-Dist: build; extra == "dev"
 Requires-Dist: twine; extra == "dev"

{proxyml-0.3.0 → proxyml-0.4.0}/src/proxyml.egg-info/requires.txt

@@ -12,4 +12,4 @@ black
 ruff
 
 [local]
-proxyml-core[modeling]>=0.1
+proxyml-core[modeling]>=0.2

{proxyml-0.3.0 → proxyml-0.4.0}/tests/test_local_challenger.py

@@ -6,6 +6,7 @@ from proxyml.local import (
     Complexity,
     LADDERS,
     TrainedChallenger,
+    score_champion,
     train_auto_challenger,
     train_challenger,
 )
@@ -139,6 +140,47 @@ def test_train_auto_challenger_matches_manual_schema_and_train():
     )
 
 
+def test_score_champion_matches_train_challenger_metric_shape_classification():
+    schema = _schema()
+    df = _df(seed=9, n=300)
+    target = np.where(df["age"] > 50, "senior", "junior")
+
+    result = train_challenger(df, target, schema, complexity=Complexity.MODERATE, task="classification")
+    champion_metrics = score_champion(target, target, task="classification")
+    assert set(champion_metrics) == set(result.metrics)
+
+
+def test_score_champion_matches_train_challenger_metric_shape_regression():
+    schema = _schema()
+    df = _df(seed=10)
+    target = df["age"] * 0.5 + df["income"] * 0.0001
+
+    result = train_challenger(df, target, schema, complexity=Complexity.MODERATE, task="regression")
+    champion_metrics = score_champion(target, target, task="regression")
+    assert set(champion_metrics) == set(result.metrics)
+
+
+def test_score_champion_uses_same_scoring_as_train_challenger():
+    # Feed score_champion the exact labels/predictions a regression fit produced
+    # internally, and assert numeric equality with the challenger's own r2 —
+    # proving the two aren't drifting copies of the same formula.
+    schema = _schema()
+    df = _df(seed=11, n=300)
+    target = df["age"] * 0.5 + df["income"] * 0.0001
+
+    result = train_challenger(df, target, schema, complexity=Complexity.MODERATE, task="regression")
+    X = df[["age", "income"]].to_numpy(dtype=object)
+    y_pred = result.pipeline.predict(X)
+    reproduced_metrics = score_champion(target, y_pred, task="regression")
+
+    # Not identical to result.metrics (that was scored on a held-out test split,
+    # this is scored on the full data) but both must use the same r2 formula —
+    # verify by comparing against sklearn directly.
+    from sklearn.metrics import r2_score
+
+    assert reproduced_metrics["r2"] == pytest.approx(r2_score(target, y_pred))
+
+
 def test_train_auto_challenger_passes_immutable_cols_to_get_schema():
     from unittest.mock import patch