proxyml 0.4.0__tar.gz → 0.5.0__tar.gz

{proxyml-0.4.0 → proxyml-0.5.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: proxyml
-Version: 0.4.0
+Version: 0.5.0
 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; extra == "local"
+Requires-Dist: proxyml-core[modeling]>=0.2.1; extra == "local"
 Provides-Extra: dev
 Requires-Dist: build; extra == "dev"
 Requires-Dist: twine; extra == "dev"

{proxyml-0.4.0 → proxyml-0.5.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "proxyml"
-version = "0.4.0"
+version = "0.5.0"
 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",
+    "proxyml-core[modeling]>=0.2.1",
 ]
 dev = [
     "build",

{proxyml-0.4.0 → proxyml-0.5.0}/src/proxyml/local/__init__.py

@@ -11,6 +11,7 @@ from proxyml.local.challenger import (
     Rung,
     TrainedChallenger,
     score_champion,
+    to_challenger_upload,
     train_auto_challenger,
     train_challenger,
 )
@@ -19,6 +20,7 @@ __all__ = [
     "train_challenger",
     "train_auto_challenger",
     "score_champion",
+    "to_challenger_upload",
     "Complexity",
     "Rung",
     "TrainedChallenger",

{proxyml-0.4.0 → proxyml-0.5.0}/src/proxyml/local/challenger.py

@@ -12,6 +12,7 @@ from __future__ import annotations
 
 from dataclasses import dataclass, replace
 from enum import Enum
+from importlib.metadata import version as _pkg_version
 from pathlib import Path
 from typing import Any, Callable, Literal
 
@@ -60,6 +61,8 @@ def _simple_classifier() -> BaseEstimator:
         max_iter=500,
         cv=5,
         n_jobs=-1,
+        scoring="accuracy",
+        use_legacy_attributes=False,
     )
 
 
@@ -76,6 +79,8 @@ def _flexible_classifier() -> BaseEstimator:
         max_iter=1000,
         cv=5,
         n_jobs=-1,
+        scoring="accuracy",
+        use_legacy_attributes=False,
     )
 
 
@@ -109,6 +114,7 @@ LADDERS: dict[Complexity, Rung] = {
 class TrainedChallenger:
     pipeline: Pipeline
     task: Literal["classification", "regression"]
+    complexity: Complexity
     metrics: dict[str, float]
     hyperparameters: dict[str, Any]
     export: SurrogateExport
@@ -192,6 +198,7 @@ def train_challenger(
     return TrainedChallenger(
         pipeline=pipeline,
         task=resolved_task,
+        complexity=complexity,
         metrics=metrics,
         hyperparameters=hyperparameters,
         export=export,
@@ -220,6 +227,63 @@ def score_champion(
     return score_predictions(np.asarray(labels), np.asarray(predictions), task=task)
 
 
+def to_challenger_upload(
+    result: TrainedChallenger,
+    *,
+    n_samples: int,
+    champion_metrics: dict[str, float] | None = None,
+    sdk_version: str | None = None,
+    proxyml_core_version: str | None = None,
+) -> dict[str, Any]:
+    """Assemble the JSON-serializable payload for a challenger upload.
+
+    Matches the shape ProxyML's dashboard/API expects at
+    ``POST /app/projects/{id}/challenger`` — handles the mechanical assembly
+    (serializing the export, stamping SDK/core versions, converting
+    ``complexity`` to a plain string) so callers don't have to hand-roll it.
+    The result is plain ``dict``/``str``/``float`` data, ready for
+    ``json.dump`` — upload it either by POSTing it directly, or by saving it
+    to a file and using the dashboard's "Upload challenger" button.
+
+    ``champion_metrics`` is optional: pass ``None`` (the default) to get a
+    self-contained export of the challenger alone — e.g. to save/share it
+    before you have a champion to compare against — and fill in
+    ``champion_metrics`` later. The upload endpoint itself still requires
+    ``champion_metrics`` at upload time; this function just doesn't force you
+    to have it up front.
+
+    Args:
+        result: output of ``train_challenger()``/``train_auto_challenger()``.
+        n_samples: size of the evaluation set both ``result.metrics`` and
+            ``champion_metrics`` were scored on. Not derived automatically —
+            ``TrainedChallenger`` doesn't retain its internal held-out split,
+            and ``champion_metrics`` typically comes from a separate
+            ``score_champion()`` call the two need to share, so the caller is
+            the only one who actually knows this number.
+        champion_metrics: the champion's real-world performance, from
+            ``score_champion()`` — same metric keys as ``result.metrics``.
+            Omit if you don't have it yet.
+        sdk_version: defaults to the installed ``proxyml`` version.
+        proxyml_core_version: defaults to the installed ``proxyml-core`` version.
+    """
+    if sdk_version is None:
+        sdk_version = _pkg_version("proxyml")
+    if proxyml_core_version is None:
+        proxyml_core_version = _pkg_version("proxyml-core")
+
+    payload: dict[str, Any] = {
+        "export": result.export.to_dict(),
+        "challenger_metrics": result.metrics,
+        "n_samples": n_samples,
+        "complexity": result.complexity.value,
+        "sdk_version": sdk_version,
+        "proxyml_core_version": proxyml_core_version,
+    }
+    if champion_metrics is not None:
+        payload["champion_metrics"] = champion_metrics
+    return payload
+
+
 def train_auto_challenger(
     data: str | Path | pd.DataFrame,
     target_col: str,

{proxyml-0.4.0 → proxyml-0.5.0}/src/proxyml.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: proxyml
-Version: 0.4.0
+Version: 0.5.0
 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; extra == "local"
+Requires-Dist: proxyml-core[modeling]>=0.2.1; extra == "local"
 Provides-Extra: dev
 Requires-Dist: build; extra == "dev"
 Requires-Dist: twine; extra == "dev"

{proxyml-0.4.0 → proxyml-0.5.0}/src/proxyml.egg-info/requires.txt

@@ -12,4 +12,4 @@ black
 ruff
 
 [local]
-proxyml-core[modeling]>=0.2
+proxyml-core[modeling]>=0.2.1

{proxyml-0.4.0 → proxyml-0.5.0}/tests/test_local_challenger.py

@@ -1,3 +1,5 @@
+import warnings
+
 import numpy as np
 import pandas as pd
 import pytest
@@ -7,6 +9,7 @@ from proxyml.local import (
     LADDERS,
     TrainedChallenger,
     score_champion,
+    to_challenger_upload,
     train_auto_challenger,
     train_challenger,
 )
@@ -63,6 +66,20 @@ def test_train_challenger_classification():
     assert result.export.classes is not None
 
 
+def test_train_challenger_classification_rungs_fit_without_sklearn_deprecation_warnings():
+    # scoring="accuracy" and use_legacy_attributes=False on the SIMPLE/FLEXIBLE
+    # classifiers pin today's sklearn defaults explicitly, matching what
+    # get_default_classifier() (the MODERATE rung) already does.
+    schema = _schema()
+    df = _df(seed=12, n=200)
+    target = np.where(df["age"] > 50, "senior", "junior")
+
+    with warnings.catch_warnings():
+        warnings.simplefilter("error", FutureWarning)
+        train_challenger(df, target, schema, complexity=Complexity.SIMPLE, task="classification")
+        train_challenger(df, target, schema, complexity=Complexity.FLEXIBLE, task="classification")
+
+
 def test_train_challenger_simple_rung_is_more_regularized_than_flexible():
     schema = _schema()
     df = _df(seed=3)
@@ -181,6 +198,94 @@ def test_score_champion_uses_same_scoring_as_train_challenger():
     assert reproduced_metrics["r2"] == pytest.approx(r2_score(target, y_pred))
 
 
+def test_train_challenger_result_carries_complexity():
+    schema = _schema()
+    df = _df(seed=13)
+    target = df["age"] * 0.5 + df["income"] * 0.0001
+
+    result = train_challenger(df, target, schema, complexity=Complexity.FLEXIBLE, task="regression")
+
+    assert result.complexity is Complexity.FLEXIBLE
+
+
+def test_train_auto_challenger_result_carries_complexity():
+    df = _labeled_df(seed=14)
+    result = train_auto_challenger(df, "approved", task="classification", complexity=Complexity.SIMPLE)
+
+    assert result.complexity is Complexity.SIMPLE
+
+
+def test_to_challenger_upload_shape_without_champion_metrics():
+    schema = _schema()
+    df = _df(seed=15)
+    target = df["age"] * 0.5 + df["income"] * 0.0001
+    result = train_challenger(df, target, schema, complexity=Complexity.MODERATE, task="regression")
+
+    payload = to_challenger_upload(result, n_samples=40)
+
+    assert payload["export"] == result.export.to_dict()
+    assert payload["challenger_metrics"] == result.metrics
+    assert payload["n_samples"] == 40
+    assert payload["complexity"] == "moderate"
+    assert "champion_metrics" not in payload
+    assert isinstance(payload["sdk_version"], str) and payload["sdk_version"]
+    assert isinstance(payload["proxyml_core_version"], str) and payload["proxyml_core_version"]
+
+
+def test_to_challenger_upload_includes_champion_metrics_when_given():
+    schema = _schema()
+    df = _df(seed=16)
+    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")
+
+    payload = to_challenger_upload(result, n_samples=40, champion_metrics=champion_metrics)
+
+    assert payload["champion_metrics"] == champion_metrics
+
+
+def test_to_challenger_upload_defaults_versions_to_installed_packages():
+    from importlib.metadata import version as pkg_version
+
+    schema = _schema()
+    df = _df(seed=17)
+    target = df["age"] * 0.5 + df["income"] * 0.0001
+    result = train_challenger(df, target, schema, task="regression")
+
+    payload = to_challenger_upload(result, n_samples=10)
+
+    assert payload["sdk_version"] == pkg_version("proxyml")
+    assert payload["proxyml_core_version"] == pkg_version("proxyml-core")
+
+
+def test_to_challenger_upload_allows_version_override():
+    schema = _schema()
+    df = _df(seed=18)
+    target = df["age"] * 0.5 + df["income"] * 0.0001
+    result = train_challenger(df, target, schema, task="regression")
+
+    payload = to_challenger_upload(
+        result, n_samples=10, sdk_version="9.9.9", proxyml_core_version="8.8.8"
+    )
+
+    assert payload["sdk_version"] == "9.9.9"
+    assert payload["proxyml_core_version"] == "8.8.8"
+
+
+def test_to_challenger_upload_payload_is_json_serializable():
+    import json
+
+    schema = _schema()
+    df = _df(seed=19, n=300)
+    target = np.where(df["age"] > 50, "senior", "junior")
+    result = train_challenger(df, target, schema, task="classification")
+    champion_metrics = score_champion(target, target, task="classification")
+
+    payload = to_challenger_upload(result, n_samples=300, champion_metrics=champion_metrics)
+
+    json.dumps(payload)  # must not raise
+
+
 def test_train_auto_challenger_passes_immutable_cols_to_get_schema():
     from unittest.mock import patch