Ruifeng Zheng created SPARK-61519:
-------------------------------------

             Summary: Support native GBTClassifier training and prediction
                 Key: SPARK-61519
                 URL: https://issues.apache.org/jira/browse/SPARK-61519
             Project: Spark
          Issue Type: New Feature
          Components: Connect
    Affects Versions: connect-rust-4.3.0
            Reporter: Ruifeng Zheng


Add native Rust GBTClassifier and GBTClassificationModel with all 22 official 
PySpark constructor parameters plus probability/raw prediction column names and 
thresholds. Expose native Python bindings with the official keyword-only 
signature and setters. Fit Spark JVM gradient-boosted binary classification 
through Connect and retain the cached model across clones and lazy transformed 
DataFrames. Validate nonlinear labels, weights, raw margins, probabilities, 
learning rate, iteration count, validation-set early stopping, subsampling, 
seed wire boundaries and model lifetime with golden fixtures and live 
Rust/native Python/reference tests. Model inspection, multiclass training, 
per-iteration evaluation, single-vector prediction and persistence are outside 
this addition.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to