callback EvaluationMonitor not working

[h4ck4l1]

import xgboost as xgb

xgb_params = {
    "objective" : "reg:absoluteerror",
    "eval_metric" :"mae",
    "seed" : 420,
    "booster" : "gbtree",
    "device" : "cuda",
    "eta" : 0.01,
    "gamma" : 5,
    "max_depth" : 128,
    "lambda" : 10,
    "alpha" : 2,
    "max_leaves": 256
}

xgb.set_config(
    verbosity=0
)

cv = GroupShuffleSplit(n_splits=5,test_size=0.1)
xgb_models_list = []
for train_ind,test_ind in cv.split(
    temp_train_df.drop('sales').to_numpy(),
    temp_train_df.get_column('sales').to_numpy(),
    groups=temp_train_df.get_column('warehouse').to_numpy()
    ):

    X_train = temp_train_df.drop('sales').with_row_index().filter(pl.col('index').is_in(train_ind)).to_numpy()
    X_test = temp_train_df.drop('sales').with_row_index().filter(pl.col('index').is_in(test_ind)).to_numpy()
    y_train = temp_train_df.select('sales').with_row_index().filter(pl.col('index').is_in(train_ind)).get_column('sales').to_numpy()
    y_test = temp_train_df.select('sales').with_row_index().filter(pl.col('index').is_in(test_ind)).get_column('sales').to_numpy()

    dtrain = xgb.DMatrix(X_train,label=y_train)
    dval = xgb.DMatrix(X_test,label=y_test)

    print(f"Training XGB model")
    xgb_model = xgb.train(
        xgb_params,
        dtrain,
        num_boost_round=4000,
        evals=[(dtrain,"Train"),(dval,"Valid")],
        callbacks=[
            xgb.callback.EarlyStopping(rounds=200),
            xgb.callback.EvaluationMonitor(period=500,show_stdv=False,rank=0)
        ]
    )

Training XGB model
[0]	Train-mae:90.99591	Valid-mae:79.83470
[0]	Train-mae:90.99591	Valid-mae:79.83470
[1]	Train-mae:90.81372	Valid-mae:79.78599
[2]	Train-mae:90.62106	Valid-mae:79.74804
[3]	Train-mae:90.44182	Valid-mae:79.70253
[4]	Train-mae:90.26843	Valid-mae:79.66690
[5]	Train-mae:90.10026	Valid-mae:79.62726
[6]	Train-mae:89.93129	Valid-mae:79.60968
[7]	Train-mae:89.76939	Valid-mae:79.57884
[8]	Train-mae:89.60516	Valid-mae:79.54222
[9]	Train-mae:89.44614	Valid-mae:79.50914
[10]	Train-mae:89.28421	Valid-mae:79.48071
[11]	Train-mae:89.11796	Valid-mae:79.45545
[12]	Train-mae:88.95689	Valid-mae:79.42221
[13]	Train-mae:88.79400	Valid-mae:79.39508
[14]	Train-mae:88.64471	Valid-mae:79.36411
[15]	Train-mae:88.48538	Valid-mae:79.33596
[16]	Train-mae:88.32891	Valid-mae:79.30990
[17]	Train-mae:88.19119	Valid-mae:79.28522

I tried using xgbRegressor class it doesn't accept callback argument and xgb.cv gives cannot convert to string to float error.

Please help!

[trivialfis]

Hi, if you are passing the EvaluationMonitor yourself, then you should set the verbose_eval=False for the train function.

I tried using xgbRegressor class it doesn't accept callback argument

It does. In the constructor:

reg = xgb.XGBRegressor(callbacks=[MyCallback()])
# or
reg.set_params(callbacks=[MyCallback()])