Unknown metric function rmsle and reg:squaredlogerror


#1

Hi,

I am using xgboost ver9.0 and went into error belows

> xgb_pars = {'min_child_weight': 50, 'eta': 0.1, 'colsample_bytree': 0.3, 'max_depth': 5,
>             'subsample': 0.8, 'lambda': 1., 'nthread': 4, 'booster' : 'gbtree', 'silent': 1,
>             'eval_metric':'rmse', 'objective': 'reg:squaredlogerror'}
> model = xgb.train(xgb_pars, dtrain, 60, watchlist, early_stopping_rounds=50,
>                   maximize=False, verbose_eval=10)


> XGBoostError: [11:46:26] src/metric/metric.cc:23: Unknown metric function rmsle
> 
> XGBoostError: [11:50:45] src/objective/objective.cc:23: Unknown objective function reg:squaredlogerror

#2

RMSLE is not part of 0.90 release. See https://xgboost.readthedocs.io/en/release_0.90/parameter.html for the parameters available for 0.90. Alternatively, you can compile the latest XGBoost source to use RMSLE.


#3

Is there any easy way to upgrade to newest release? I am using conda.

Thanks


#4

You can download wheels from nightly builds: