classification xgboost train bug

[shenzhengntu]

\set ON_ERROR_STOP true
\timing on

-- 清理旧数据
DROP EXTENSION IF EXISTS pgml CASCADE;
DROP SCHEMA IF EXISTS pgml CASCADE;
CREATE EXTENSION pgml;

-- 加载数据集
SELECT pgml.load_dataset('breast_cancer');
-- 加载数据集
SELECT pgml.load_dataset('breast_cancer');

-- =============================================
-- 分类算法 Python vs Rust 对比测试
-- 指标: f1, 差异阈值: < 0.05 (random_forest < 0.1)
-- =============================================

-- === linear 分类 ===
SELECT * FROM pgml.train('Rust Classification Compare', 'classification', 'pgml.breast_cancer', 'malignant', algorithm => 'linear', runtime => 'python'); SELECT * FROM pgml.train('Rust Classification Compare', algorithm => 'linear', runtime => 'rust'); -- === xgboost 分类 ===
SELECT * FROM pgml.train('Rust Classification Compare', algorithm => 'xgboost', runtime => 'python', hyperparams => '{"n_estimators": 10}'); SELECT * FROM pgml.train('Rust Classification Compare', algorithm => 'xgboost', runtime => 'rust', hyperparams => '{"n_estimators": 10}');

-- xgboost rust "roc_auc" always 0.5-----

INFO: Metrics: {"roc_auc": 0.5, "log_loss": 10.55341, "f1": 0.50526315, "precision": 0.33802816, "recall": 1.0, "accuracy": 0.33802816, "mcc": NaN, "fit_time": 0.006182069, "score_time": 0.000807744} INFO: Comparing to deployed model f1: Some(0.9292929172515868) WARNING: New model's f1 is not better than current model. New: 0.5052631497383118, Current 0.9292929172515868 WARNING: Not deploying newly trained model.
project | task | algorithm | deployed
-----------------------------+----------------+-----------+----------
Rust Classification Compare | classification | xgboost | f

the reason is not update bst train ,lead to "roc_auc": 0.5,the normal value should be around 0.9.