Ok, I finally solved it - added a column to the dataset validation_indicator_col='validation_0', and did not pass it the the VectorAssembler:

xgboost_regressor = SparkXGBRegressor()
xgboost_regressor.setParams(
    gamma=0.2,
    max_depth=6,
    objective="reg:logistic",       # logistic regression, output probability
    missing=MISSING_VALUE_NUM_DEFAULT,
    num_workers=60,
    subsample=0.5,
    colsample_bytree=0.7,
    learning_rate=0.01,
    random_state=1234,
    reg_alpha=0.35,
    reg_lambda=0.3,
    n_estimators=600,
    eval_metric='auc',
    early_stopping_rounds=5,
    validation_indicator_col='validation_0',
    maximize=True,
    verbose=True,
    )