Reproduction Steps:

1. Start the official container using the following command:
   docker run \ -it \ -v postgresml_data:/var/lib/postgresql \ -p 5433:5432 \ -p 8000:8000 \ ghcr.io/postgresml/postgresml:2.9.3 \ sudo -u postgresml psql -d postgresml

Issue 1: F1 Score is NaN
After executing the following SQL commands:
SELECT pgml.load_dataset('digits');
SELECT * FROM pgml.train('Handwritten Digits', 'classification', 'pgml.digits', 'target', algorithm => 'xgboost', hyperparams => '{"n_estimators": 10}');
The output:

The issue arises where the recall becomes NaN. This is due to the model trained by XGBoost classifying all test instances into the category "9," leading to TP + FN equals 0, hence causing division by zero in F1 score calculation. Notably, this issue does not surface when using runtime => python.

Issue 2: Training Error When Assigning "objective" Hyperparameter
Attempting to specify the "objective" hyperparam during training leads to an error.
SELECT * FROM pgml.train('Handwritten Digits', 'classification', 'pgml.digits', 'target', algorithm => 'xgboost', hyperparams => '{"n_estimators": 10, "objective": "multi:softmax"}');
The error:

Similarly, there are no issues with the Python XGBoost.