spforge 0.8.4__py3-none-any.whl → 0.8.7__py3-none-any.whl

{spforge-0.8.4.dist-info → spforge-0.8.7.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: spforge
-Version: 0.8.4
+Version: 0.8.7
 Summary: A flexible framework for generating features, ratings, and building machine learning or other models for training and inference on sports data.
 Author-email: Mathias Holmstrøm <mathiasholmstom@gmail.com>
 License: See LICENSE file
@@ -17,7 +17,7 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: numpy>=1.23.4
 Requires-Dist: optuna>=3.4.0
-Requires-Dist: pandas>=2.0.0
+Requires-Dist: pandas<3.0.0,>=2.0.0
 Requires-Dist: pendulum>=1.0.0
 Requires-Dist: scikit-learn>=1.4.0
 Requires-Dist: lightgbm>=4.0.0
@@ -85,12 +85,12 @@ This example demonstrates predicting NBA game winners using player-level ratings
 import pandas as pd
 from sklearn.linear_model import LogisticRegression
 
+from examples import get_sub_sample_nba_data
 from spforge.autopipeline import AutoPipeline
 from spforge.data_structures import ColumnNames
-from spforge.ratings import RatingKnownFeatures
-from spforge.ratings._player_rating import PlayerRatingGenerator
+from spforge.ratings import PlayerRatingGenerator, RatingKnownFeatures
 
-df = pd.read_parquet("data/game_player_subsample.parquet")
+df = get_sub_sample_nba_data(as_pandas=True, as_polars=False)
 
 # Step 1: Define column mappings for your dataset
 column_names = ColumnNames(
@@ -144,7 +144,7 @@ historical_df = rating_generator.fit_transform(historical_df)
 pipeline = AutoPipeline(
     estimator=LogisticRegression(),
     granularity=["game_id", "team_id"],  # Aggregate players → teams
-    feature_names=rating_generator.features_out + ["location"],  # Rating + home/away
+    estimator_features=rating_generator.features_out + ["location"],  # Rating + home/away
 )
 
 # Train on historical data
@@ -302,8 +302,8 @@ cross_validator = MatchKFoldCrossValidator(
     prediction_column_name="points_pred",
     target_column="points",
     n_splits=3,  # Number of temporal folds
-    # Must include both feature_names AND context_feature_names
-    features=pipeline.feature_names + pipeline.context_feature_names,
+    # Must include both estimator features and context features
+    features=pipeline.required_features,
 )
 
 # Generate validation predictions
@@ -330,7 +330,7 @@ print(f"Validation MAE: {mae:.2f}")
   - `is_validation=1` marks validation rows, `is_validation=0` marks training rows
   - Use `validation_column` in scorer to score only validation rows
 - Training data always comes BEFORE validation data chronologically
-- Must pass both `feature_names` + `context_feature_names` to `features` parameter
+- Must pass all required features (use `pipeline.required_features`)
 - Scorers can filter rows (e.g., only score players who played minutes > 0)
 
 See [examples/nba/cross_validation_example.py](examples/nba/cross_validation_example.py) for a complete example.
@@ -371,7 +371,7 @@ from lightgbm import LGBMClassifier, LGBMRegressor
 # Approach 1: LGBMClassifier (direct probability prediction)
 pipeline_classifier = AutoPipeline(
     estimator=LGBMClassifier(verbose=-100, random_state=42),
-    feature_names=features_pipeline.features_out,
+    estimator_features=features_pipeline.features_out,
 )
 
 # Approach 2: LGBMRegressor + NegativeBinomialEstimator
@@ -385,13 +385,7 @@ distribution_estimator = NegativeBinomialEstimator(
 
 pipeline_negbin = AutoPipeline(
     estimator=distribution_estimator,
-    feature_names=features_pipeline.features_out,
-    context_feature_names=[
-        column_names.player_id,
-        column_names.start_date,
-        column_names.team_id,
-        column_names.match_id,
-    ],
+    estimator_features=features_pipeline.features_out,
     predictor_transformers=[
         EstimatorTransformer(
             prediction_column_name="points_estimate",
@@ -439,7 +433,7 @@ points_estimate_transformer = EstimatorTransformer(
 # Stage 2: Refine estimate using Stage 1 output
 player_points_pipeline = AutoPipeline(
     estimator=LGBMRegressor(verbose=-100, n_estimators=50),
-    feature_names=features_pipeline.features_out,  # Original features
+    estimator_features=features_pipeline.features_out,  # Original features
     # predictor_transformers execute first, adding their predictions
     predictor_transformers=[points_estimate_transformer],
 )
@@ -474,4 +468,3 @@ For complete, runnable examples with detailed explanations:
 - **[examples/nba/cross_validation_example.py](examples/nba/cross_validation_example.py)** - Time-series CV, distributions, and scoring
 - **[examples/nba/predictor_transformers_example.py](examples/nba/predictor_transformers_example.py)** - Multi-stage hierarchical modeling
 - **[examples/nba/game_winner_example.py](examples/nba/game_winner_example.py)** - Basic workflow for game winner prediction
-

{spforge-0.8.4.dist-info → spforge-0.8.7.dist-info}/RECORD

@@ -1,19 +1,19 @@
 examples/__init__.py,sha256=qGLpphvrjQj0-zS9vP0Q07L-anDnmw7gFZJUEBgYG3U,158
 examples/game_level_example.py,sha256=EOr-H0K79O3Zah4wWuqa5DLmT2iZGbfgxD-xSU2-dfI,2244
 examples/lol/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-examples/lol/pipeline_transformer_example.py,sha256=HousFjE2dbJgdONur4PxwhW2SGQIJGI8aZUIb4TEvIo,4317
+examples/lol/pipeline_transformer_example.py,sha256=XVmm6Xya5z7JyOA0s-DISOlR2I1wpUthCyhRSt9n6qE,3402
 examples/lol/data/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 examples/lol/data/subsample_lol_data.parquet,sha256=tl04XDslylECJUV1e0DGeqMb6D0Uh6_48NO6TykdgQI,343549
 examples/lol/data/utils.py,sha256=Lt3XNNa5cavvFXHaTQ-GOPxSuWmPEfEO0CVXQEyF_s0,486
 examples/nba/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-examples/nba/cross_validation_example.py,sha256=WD_52vO9m1rILVfXXf8uIb_odpaK-TZ4iOewHU19lTg,5281
-examples/nba/feature_engineering_example.py,sha256=0OHJ2w6vkHvFB2bYwIQQb8HjFA5bfXc7tLmngnahG74,7708
-examples/nba/game_winner_example.py,sha256=RNKYSwpArr08yDWOtkxjx7eAldf97WYDBBfb3tsVSZc,2975
-examples/nba/predictor_transformers_example.py,sha256=mPXRVPx4J5VZtxYH89k7pwh7_EGZ0CXoNHeh2s0AOp8,8499
+examples/nba/cross_validation_example.py,sha256=XVnQJ5mqMou9z83ML5J0wS3gk-pa56sdvahJYQgZ8os,5056
+examples/nba/feature_engineering_example.py,sha256=BDd5594Yi_56lGDqz3SYQkwT8NVZyFkgv3gKPCsAjz4,8197
+examples/nba/game_winner_example.py,sha256=7VVHxGyU2uPjT9q6lDMHJ5KpkWp9gU8brxr_UZfuSHg,3189
+examples/nba/predictor_transformers_example.py,sha256=Fl4BY_hVW0iYERolN6s-ZB2xv-UxOK547L6iI5t0r0Y,8807
 examples/nba/data/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 examples/nba/data/game_player_subsample.parquet,sha256=ODJxHC-mUYbJ7r-ScUFtPU7hrFuxLUbbDSobmpCkw0w,279161
 examples/nba/data/utils.py,sha256=41hxLQ1d6ZgBEcHa5MI0-fG5KbsRi07cclMPQZM95ek,509
-spforge/__init__.py,sha256=5d9zzBxaaXj2JeBNwfUwuV7Ll5FERHyXONsFiuKhHSQ,402
+spforge/__init__.py,sha256=8vZhy7XUpzqWkVKpXqwqOLDkQlNytRhyf4qjwObfXgU,468
 spforge/autopipeline.py,sha256=ZUwv6Q6O8cD0u5TiSqG6lhW0j16RlSb160AzuOeL2R8,23186
 spforge/base_feature_generator.py,sha256=RbD00N6oLCQQcEb_VF5wbwZztl-X8k9B0Wlaj9Os1iU,668
 spforge/data_structures.py,sha256=k82v5r79vl0_FAVvsxVF9Nbzb5FoHqVrlHZlEXGc5gQ,7298
@@ -43,18 +43,19 @@ spforge/feature_generator/_rolling_mean_binary.py,sha256=lmODy-o9Dd9pb8IlA7g4UyA
 spforge/feature_generator/_rolling_mean_days.py,sha256=EZQmFmYVQB-JjZV5k8bOWnaTxNpPDCZAjdfdhiiG4r4,8415
 spforge/feature_generator/_rolling_window.py,sha256=HT8LezsRIPNAlMEoP9oTPW2bKFu55ZSRnQZGST7fncw,8836
 spforge/feature_generator/_utils.py,sha256=KDn33ia1OYJTK8THFpvc_uRiH_Bl3fImGqqbfzs0YA4,9654
-spforge/hyperparameter_tuning/__init__.py,sha256=pp7aWzydObRawFLcGiaUrUduEQIjln2uif9nKCTk6l4,509
-spforge/hyperparameter_tuning/_default_search_spaces.py,sha256=19sHW8zlyG88xZdyqSrp9gFI5oLb-f6THlbhYAtTfmY,3534
-spforge/hyperparameter_tuning/_tuner.py,sha256=S70IEmHxl36LaUPl_wc_2mo46qUuH8t0eH0aXuCuGfA,9586
+spforge/hyperparameter_tuning/__init__.py,sha256=N2sKG4SvG41hlsFT2kx_DQYMmXsQr-8031Tu_rxlxyY,1015
+spforge/hyperparameter_tuning/_default_search_spaces.py,sha256=Sm5IrHAW0-vRC8jqCPX0pDi_C-W3L_MoEKGA8bx1Zbc,7546
+spforge/hyperparameter_tuning/_tuner.py,sha256=uovhGqhe8-fdhi79aErUmE2h5NCycFQEIRv5WCjpC7E,16732
 spforge/performance_transformers/__init__.py,sha256=U6d7_kltbUMLYCGBk4QAFVPJTxXD3etD9qUftV-O3q4,422
 spforge/performance_transformers/_performance_manager.py,sha256=KwAga6dGhNkXi-MDW6LPjwk6VZwCcjo5L--jnk9aio8,9706
 spforge/performance_transformers/_performances_transformers.py,sha256=0lxuWjAfWBRXRgQsNJHjw3P-nlTtHBu4_bOVdoy7hq4,15536
-spforge/ratings/__init__.py,sha256=jAa_xF2e-96FoyD57EYFKE-mO6OnK23siJOB4tzbyek,387
+spforge/ratings/__init__.py,sha256=OZVH2Lo6END3n1X8qi4QcyAPlThIwAYwVKCiIuOQSQU,576
 spforge/ratings/_base.py,sha256=dRMkIGj5-2zKddygaEA4g16WCyXon7v8Xa1ymm7IuoM,14335
-spforge/ratings/_player_rating.py,sha256=05CuiSa2_uM0xtYpxT00OOxU_TmW4qt6dsXvn7seFss,50861
+spforge/ratings/_player_rating.py,sha256=MyqsyLSY6d7_bxDSnF8eWOyXpSCADWGdepdFSGM4cHw,51365
 spforge/ratings/_team_rating.py,sha256=T0kFiv3ykYSrVGGsVRa8ZxLB0WMnagxqdFDzl9yZ_9g,24813
 spforge/ratings/enums.py,sha256=s7z_RcZS6Nlgfa_6tasO8_IABZJwywexe7sep9DJBgo,1739
 spforge/ratings/league_identifier.py,sha256=_KDUKOwoNU6RNFKE5jju4eYFGVNGBdJsv5mhNvMakfc,6019
+spforge/ratings/league_start_rating_optimizer.py,sha256=Q4Vo3QT-r55qP4aD9WftsTB00UOSRvxM1khlyuAGWNM,8582
 spforge/ratings/player_performance_predictor.py,sha256=cMxzQuk0nF1MsT_M32g-3mxVdAEbZ-S7TUjEPYdo3Yg,8361
 spforge/ratings/start_rating_generator.py,sha256=_7hIJ9KRVCwsCoY1GIzY8cuOdHR8RH_BCMeMwQG3E04,6776
 spforge/ratings/team_performance_predictor.py,sha256=ThQOmYQUqKBB46ONYHOMM2arXFH8AkyKpAZzs80SjHA,7217
@@ -70,15 +71,17 @@ spforge/transformers/_other_transformer.py,sha256=xLfaFIhkFsigAoitB4x3F8An2j9ymd
 spforge/transformers/_predictor.py,sha256=2sE6gfVrilXzPVcBurSrtqHw33v2ljygQcEYXt9LhZc,3119
 spforge/transformers/_simple_transformer.py,sha256=zGUFNQYMeoDSa2CoQejQNiNmKCBN5amWTvyOchiUHj0,5660
 spforge/transformers/_team_ratio_predictor.py,sha256=g8_bR53Yyv0iNCtol1O9bgJSeZcIco_AfbQuUxQJkeY,6884
-spforge-0.8.4.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+spforge-0.8.7.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
 tests/test_autopipeline.py,sha256=WXHeqBdjQD6xaXVkzvS8ocz0WVP9R7lN0PiHJ2iD8nA,16911
 tests/test_autopipeline_context.py,sha256=IuRUY4IA6uMObvbl2pXSaXO2_tl3qX6wEbTZY0dkTMI,1240
 tests/test_feature_generator_pipeline.py,sha256=CAgBknWqawqYi5_hxcPmpxrLVa5elMHVv1VrSVRKXEA,17705
 tests/cross_validator/test_cross_validator.py,sha256=itCGhNY8-NbDbKbhxHW20wiLuRst7-Rixpmi3FSKQtA,17474
 tests/distributions/test_distribution.py,sha256=aU8hfCgliM80TES4WGjs9KFXpV8XghBGF7Hu9sqEVSE,10982
+tests/end_to_end/test_estimator_hyperparameter_tuning.py,sha256=fZCJ9rrED2vT68B9ovmVA1cIG2pHRTjy9xzZLxxpEBo,2513
+tests/end_to_end/test_league_start_rating_optimizer.py,sha256=Mmct2ixp4c6L7PGym8wZc7E-Csozryt1g4_o6OCc1uI,3141
 tests/end_to_end/test_lol_player_kills.py,sha256=RJSYUbPrZ-RzSxGggj03yN0JKYeTB1JghVGYFMYia3Y,11891
 tests/end_to_end/test_nba_player_points.py,sha256=kyzjo7QIcvpteps29Wix6IS_eJG9d1gHLeWtIHpkWMs,9066
-tests/end_to_end/test_nba_player_ratings_hyperparameter_tuning.py,sha256=eOsTSVWv16bc0l_nCxH4x8jF-gsmn4Ttfv92mHqSXzc,6303
+tests/end_to_end/test_nba_player_ratings_hyperparameter_tuning.py,sha256=LXRkI_6Ho2kzJVbNAM17QFhx_MP9WdDJXCO9dWgJGNA,6491
 tests/end_to_end/test_nba_prediction_consistency.py,sha256=o3DckJasx_I1ed6MhMYZUo2WSDvQ_p3HtJa9DCWTIYU,9857
 tests/estimator/test_sklearn_estimator.py,sha256=tVfOP9Wx-tV1b6DcHbGxQHZQzNPA0Iobq8jTcUrk59U,48668
 tests/feature_generator/test_lag.py,sha256=5Ffrv0V9cwkbkzRMPBe3_c_YNW-W2al-XH_acQIvdeg,19531
@@ -87,13 +90,14 @@ tests/feature_generator/test_rolling_against_opponent.py,sha256=20kH1INrWy6DV7AS
 tests/feature_generator/test_rolling_mean_binary.py,sha256=KuIavJ37Pt8icAb50B23lxdWEPVSHQ7NZHisD1BDpmU,16216
 tests/feature_generator/test_rolling_mean_days.py,sha256=EyOvdJDnmgPfe13uQBOkwo7fAteBQx-tnyuGM4ng2T8,18884
 tests/feature_generator/test_rolling_window.py,sha256=YBJo36OK3ILYeXrH06ylXqviUcCaGYaVQaK5RJzwM7Y,23239
+tests/hyperparameter_tuning/test_estimator_tuner.py,sha256=iewME41d6LR2aQ0OtohGFtN_ocJUwTeqvs6L0QDmfG4,4413
 tests/hyperparameter_tuning/test_rating_tuner.py,sha256=PyCFP3KPc4Iy9E_X9stCVxra14uMgC1tuRwuQ30rO_o,13195
 tests/performance_transformers/test_performance_manager.py,sha256=bfC5GiBuzHw-mLmKeEzBUUPuKm0ayax2bsF1j88W8L0,10120
 tests/performance_transformers/test_performances_transformers.py,sha256=A-tGiCx7kXrj1cVj03Bc7prOeZ1_Ryz8YFx9uj3eK6w,11064
-tests/ratings/test_player_rating_generator.py,sha256=3mjqlX159QqOlBoY3r_TFkvLwpE4zlLE0fiqpbfk3ps,58547
+tests/ratings/test_player_rating_generator.py,sha256=FGH3Tq0uFoSlkS_XMldsUKhsovBRBvzH9EbqjKvg2O0,59601
 tests/ratings/test_ratings_property.py,sha256=ckyfGILXa4tfQvsgyXEzBDNr2DUmHwFRV13N60w66iE,6561
 tests/ratings/test_team_rating_generator.py,sha256=cDnf1zHiYC7pkgydE3MYr8wSTJIq-bPfSqhIRI_4Tic,95357
-tests/scorer/test_score.py,sha256=KTrGJypQEpU8tmgJ6LU8wK1SRC3PLUXFzZIyiA-UY7U,71749
+tests/scorer/test_score.py,sha256=_Vd6tKpy_1GeOxU7Omxci4CFf7PvRGMefEI0gv2gV6A,74688
 tests/scorer/test_score_aggregation_granularity.py,sha256=h-hyFOLzwp-92hYVU7CwvlRJ8jhB4DzXCtqgI-zcoqM,13677
 tests/transformers/test_estimator_transformer_context.py,sha256=5GOHbuWCWBMFwwOTJOuD4oNDsv-qDR0OxNZYGGuMdag,1819
 tests/transformers/test_net_over_predicted.py,sha256=vh7O1iRRPf4vcW9aLhOMAOyatfM5ZnLsQBKNAYsR3SU,3363
@@ -101,7 +105,7 @@ tests/transformers/test_other_transformer.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRk
 tests/transformers/test_predictor_transformer.py,sha256=N1aBYLjN3ldpYZLwjih_gTFYSMitrZu-PNK78W6RHaQ,6877
 tests/transformers/test_simple_transformer.py,sha256=wWR0qjLb_uS4HXrJgGdiqugOY1X7kwd1_OPS02IT2b8,4676
 tests/transformers/test_team_ratio_predictor.py,sha256=fOUP_JvNJi-3kom3ZOs1EdG0I6Z8hpLpYKNHu1eWtOw,8562
-spforge-0.8.4.dist-info/METADATA,sha256=XNaD0lL_puuuYmZU59VjenOYpLSRCSx_nswef8yCZ4M,20219
-spforge-0.8.4.dist-info/WHEEL,sha256=qELbo2s1Yzl39ZmrAibXA2jjPLUYfnVhUNTlyF1rq0Y,92
-spforge-0.8.4.dist-info/top_level.txt,sha256=6UW2M5a7WKOeaAi900qQmRKNj5-HZzE8-eUD9Y9LTq0,23
-spforge-0.8.4.dist-info/RECORD,,
+spforge-0.8.7.dist-info/METADATA,sha256=7vwprmmFvSpEL3lC0HqFZPbzxMi8mRzI0yOsa7pUlNQ,20047
+spforge-0.8.7.dist-info/WHEEL,sha256=qELbo2s1Yzl39ZmrAibXA2jjPLUYfnVhUNTlyF1rq0Y,92
+spforge-0.8.7.dist-info/top_level.txt,sha256=6UW2M5a7WKOeaAi900qQmRKNj5-HZzE8-eUD9Y9LTq0,23
+spforge-0.8.7.dist-info/RECORD,,

tests/end_to_end/test_estimator_hyperparameter_tuning.py

@@ -0,0 +1,85 @@
+import polars as pl
+from sklearn.linear_model import LogisticRegression
+from sklearn.metrics import mean_absolute_error
+
+from examples import get_sub_sample_nba_data
+from spforge import AutoPipeline, ColumnNames, EstimatorHyperparameterTuner, ParamSpec
+from spforge.cross_validator import MatchKFoldCrossValidator
+from spforge.scorer import SklearnScorer
+
+
+def test_nba_estimator_hyperparameter_tuning__workflow_completes():
+    df = get_sub_sample_nba_data(as_polars=True, as_pandas=False)
+    column_names = ColumnNames(
+        team_id="team_id",
+        match_id="game_id",
+        start_date="start_date",
+        player_id="player_id",
+        participation_weight="minutes_ratio",
+    )
+
+    df = df.sort(
+        [
+            column_names.start_date,
+            column_names.match_id,
+            column_names.team_id,
+            column_names.player_id,
+        ]
+    )
+
+    df = df.with_columns(
+        [
+            (pl.col("minutes") / pl.col("minutes").sum().over("game_id")).alias(
+                "minutes_ratio"
+            ),
+            (pl.col("points") > pl.lit(10)).cast(pl.Int64).alias("points_over_10"),
+        ]
+    )
+
+    estimator = AutoPipeline(
+        estimator=LogisticRegression(max_iter=200),
+        estimator_features=["minutes", "minutes_ratio"],
+    )
+
+    cv = MatchKFoldCrossValidator(
+        match_id_column_name=column_names.match_id,
+        date_column_name=column_names.start_date,
+        target_column="points_over_10",
+        estimator=estimator,
+        prediction_column_name="points_pred",
+        n_splits=2,
+        features=estimator.required_features,
+    )
+
+    scorer = SklearnScorer(
+        scorer_function=mean_absolute_error,
+        pred_column="points_pred",
+        target="points_over_10",
+        validation_column="is_validation",
+    )
+
+    tuner = EstimatorHyperparameterTuner(
+        estimator=estimator,
+        cross_validator=cv,
+        scorer=scorer,
+        direction="minimize",
+        param_search_space={
+            "C": ParamSpec(
+                param_type="float",
+                low=0.1,
+                high=2.0,
+                log=True,
+            ),
+        },
+        n_trials=3,
+        show_progress_bar=False,
+    )
+
+    result = tuner.optimize(df)
+
+    assert result.best_params is not None
+    assert isinstance(result.best_params, dict)
+    assert "estimator__C" in result.best_params
+    assert isinstance(result.best_value, float)
+    assert result.best_trial is not None
+    assert result.study is not None

tests/end_to_end/test_league_start_rating_optimizer.py

@@ -0,0 +1,117 @@
+import pandas as pd
+import polars as pl
+import pytest
+
+from spforge import ColumnNames
+from spforge.ratings import (
+    LeagueStartRatingOptimizer,
+    PlayerRatingGenerator,
+    TeamRatingGenerator,
+)
+
+
+def _player_df():
+    dates = pd.date_range("2024-01-01", periods=3, freq="D")
+    rows = []
+    for i, date in enumerate(dates):
+        mid = f"M{i}"
+        for player_idx in range(2):
+            rows.append(
+                {
+                    "pid": f"A{player_idx}",
+                    "tid": "TA",
+                    "mid": mid,
+                    "date": date,
+                    "league": "LCK",
+                    "perf": 0.4,
+                }
+            )
+        for player_idx in range(2):
+            rows.append(
+                {
+                    "pid": f"B{player_idx}",
+                    "tid": "TB",
+                    "mid": mid,
+                    "date": date,
+                    "league": "LEC",
+                    "perf": 0.6,
+                }
+            )
+    return pd.DataFrame(rows)
+
+
+def _team_df():
+    dates = pd.date_range("2024-01-01", periods=3, freq="D")
+    rows = []
+    for i, date in enumerate(dates):
+        mid = f"M{i}"
+        rows.extend(
+            [
+                {
+                    "tid": "TA",
+                    "mid": mid,
+                    "date": date,
+                    "league": "LCK",
+                    "perf": 0.4,
+                },
+                {
+                    "tid": "TB",
+                    "mid": mid,
+                    "date": date,
+                    "league": "LEC",
+                    "perf": 0.6,
+                },
+            ]
+        )
+    return pd.DataFrame(rows)
+
+
+@pytest.mark.parametrize("use_polars", [False, True])
+def test_league_start_rating_optimizer__adjusts_player_leagues(use_polars):
+    cn = ColumnNames(
+        player_id="pid",
+        team_id="tid",
+        match_id="mid",
+        start_date="date",
+        league="league",
+    )
+    df = _player_df()
+    if use_polars:
+        df = pl.from_pandas(df)
+    generator = PlayerRatingGenerator(performance_column="perf", column_names=cn)
+    optimizer = LeagueStartRatingOptimizer(
+        rating_generator=generator,
+        n_iterations=1,
+        learning_rate=0.5,
+        min_cross_region_rows=1,
+    )
+
+    result = optimizer.optimize(df)
+
+    assert result.league_ratings["LCK"] < 1000
+    assert result.league_ratings["LEC"] > 1000
+
+
+@pytest.mark.parametrize("use_polars", [False, True])
+def test_league_start_rating_optimizer__adjusts_team_leagues(use_polars):
+    cn = ColumnNames(
+        team_id="tid",
+        match_id="mid",
+        start_date="date",
+        league="league",
+    )
+    df = _team_df()
+    if use_polars:
+        df = pl.from_pandas(df)
+    generator = TeamRatingGenerator(performance_column="perf", column_names=cn)
+    optimizer = LeagueStartRatingOptimizer(
+        rating_generator=generator,
+        n_iterations=1,
+        learning_rate=0.5,
+        min_cross_region_rows=1,
+    )
+
+    result = optimizer.optimize(df)
+
+    assert result.league_ratings["LCK"] < 1000
+    assert result.league_ratings["LEC"] > 1000

tests/end_to_end/test_nba_player_ratings_hyperparameter_tuning.py

@@ -97,6 +97,11 @@ def test_nba_player_ratings_hyperparameter_tuning__workflow_completes(
         "confidence_max_sum",
         "use_off_def_split",
         "performance_predictor",
+        "start_team_weight",
+        "start_league_quantile",
+        "start_min_count_for_percentiles",
+        "start_min_match_count_team_rating",
+        "start_team_rating_subtract",
     }
     assert set(result.best_params.keys()) == expected_params
 

tests/hyperparameter_tuning/test_estimator_tuner.py

@@ -0,0 +1,167 @@
+import numpy as np
+import pandas as pd
+import pytest
+from sklearn.base import BaseEstimator
+from sklearn.linear_model import LogisticRegression
+
+from spforge import EstimatorHyperparameterTuner, ParamSpec
+from spforge.cross_validator import MatchKFoldCrossValidator
+from spforge.estimator import SkLearnEnhancerEstimator
+from spforge.scorer import MeanBiasScorer
+
+
+class FakeLGBMClassifier(BaseEstimator):
+    __module__ = "lightgbm.sklearn"
+
+    def __init__(
+        self,
+        n_estimators: int = 100,
+        num_leaves: int = 31,
+        max_depth: int = 5,
+        min_child_samples: int = 20,
+        subsample: float = 1.0,
+        subsample_freq: int = 1,
+        reg_alpha: float = 0.0,
+        reg_lambda: float = 0.0,
+    ):
+        self.n_estimators = n_estimators
+        self.num_leaves = num_leaves
+        self.max_depth = max_depth
+        self.min_child_samples = min_child_samples
+        self.subsample = subsample
+        self.subsample_freq = subsample_freq
+        self.reg_alpha = reg_alpha
+        self.reg_lambda = reg_lambda
+
+    def fit(self, X, y):
+        self.classes_ = np.unique(y)
+        return self
+
+    def predict_proba(self, X):
+        n = len(X)
+        if len(self.classes_) < 2:
+            return np.ones((n, 1))
+        return np.tile([0.4, 0.6], (n, 1))
+
+    def predict(self, X):
+        n = len(X)
+        if len(self.classes_) == 1:
+            return np.full(n, self.classes_[0])
+        proba = self.predict_proba(X)
+        idx = np.argmax(proba, axis=1)
+        return np.array(self.classes_)[idx]
+
+
+@pytest.fixture
+def sample_df():
+    dates = pd.date_range("2024-01-01", periods=12, freq="D")
+    rows = []
+    for i, date in enumerate(dates):
+        rows.append(
+            {
+                "mid": f"M{i // 2}",
+                "date": date,
+                "x1": float(i),
+                "y": 1 if i % 2 == 0 else 0,
+            }
+        )
+    return pd.DataFrame(rows)
+
+
+@pytest.fixture
+def scorer():
+    return MeanBiasScorer(
+        pred_column="y_pred",
+        target="y",
+        validation_column="is_validation",
+    )
+
+
+def test_estimator_tuner_requires_search_space(sample_df, scorer):
+    estimator = LogisticRegression()
+
+    cv = MatchKFoldCrossValidator(
+        match_id_column_name="mid",
+        date_column_name="date",
+        target_column="y",
+        estimator=estimator,
+        prediction_column_name="y_pred",
+        n_splits=2,
+        features=["x1"],
+    )
+
+    tuner = EstimatorHyperparameterTuner(
+        estimator=estimator,
+        cross_validator=cv,
+        scorer=scorer,
+        direction="minimize",
+        n_trials=2,
+        show_progress_bar=False,
+    )
+
+    with pytest.raises(ValueError, match="param_search_space is required"):
+        tuner.optimize(sample_df)
+
+
+def test_estimator_tuner_custom_search_space(sample_df, scorer):
+    estimator = SkLearnEnhancerEstimator(estimator=LogisticRegression())
+
+    cv = MatchKFoldCrossValidator(
+        match_id_column_name="mid",
+        date_column_name="date",
+        target_column="y",
+        estimator=estimator,
+        prediction_column_name="y_pred",
+        n_splits=2,
+        features=["x1"],
+    )
+
+    tuner = EstimatorHyperparameterTuner(
+        estimator=estimator,
+        cross_validator=cv,
+        scorer=scorer,
+        direction="minimize",
+        param_search_space={
+            "C": ParamSpec(
+                param_type="float",
+                low=0.1,
+                high=2.0,
+                log=True,
+            )
+        },
+        n_trials=2,
+        show_progress_bar=False,
+    )
+
+    result = tuner.optimize(sample_df)
+
+    assert "estimator__C" in result.best_params
+    assert isinstance(result.best_value, float)
+
+
+def test_estimator_tuner_lgbm_defaults(sample_df, scorer):
+    estimator = FakeLGBMClassifier()
+
+    cv = MatchKFoldCrossValidator(
+        match_id_column_name="mid",
+        date_column_name="date",
+        target_column="y",
+        estimator=estimator,
+        prediction_column_name="y_pred",
+        n_splits=2,
+        features=["x1"],
+    )
+
+    tuner = EstimatorHyperparameterTuner(
+        estimator=estimator,
+        cross_validator=cv,
+        scorer=scorer,
+        direction="minimize",
+        n_trials=2,
+        show_progress_bar=False,
+    )
+
+    result = tuner.optimize(sample_df)
+
+    assert "n_estimators" in result.best_params
+    assert isinstance(result.best_value, float)

tests/ratings/test_player_rating_generator.py

@@ -1662,3 +1662,30 @@ def test_player_rating_team_with_strong_offense_and_weak_defense_gets_expected_r
 
     assert a_off > start_rating
     assert a_def < start_rating
+
+
+def test_fit_transform__player_rating_difference_from_team_projected_feature(base_cn, sample_df):
+    """PLAYER_RATING_DIFFERENCE_FROM_TEAM_PROJECTED computes player_off_rating - team_off_rating_projected."""
+    gen = PlayerRatingGenerator(
+        performance_column="perf",
+        column_names=base_cn,
+        auto_scale_performance=True,
+        features_out=[
+            RatingKnownFeatures.PLAYER_RATING_DIFFERENCE_FROM_TEAM_PROJECTED,
+            RatingKnownFeatures.PLAYER_OFF_RATING,
+            RatingKnownFeatures.TEAM_OFF_RATING_PROJECTED,
+        ],
+    )
+    result = gen.fit_transform(sample_df)
+
+    diff_col = "player_rating_difference_from_team_projected_perf"
+    player_col = "player_off_rating_perf"
+    team_col = "team_off_rating_projected_perf"
+
+    assert diff_col in result.columns
+    assert player_col in result.columns
+    assert team_col in result.columns
+
+    for row in result.iter_rows(named=True):
+        expected = row[player_col] - row[team_col]
+        assert row[diff_col] == pytest.approx(expected, rel=1e-9)