0

我用 sklearn 库创建了一个模型。为了使用 dask 工作人员，我正在尝试使用 dask_ml 库创建相同的东西。下面是我的代码

from dask_ml.xgboost import XGBRegressor
from dask_ml.model_selection import RandomizedSearchCV

client = Client(os.environ['dask_scheduler'], timeout=10000)   
parameters = {
    'n_estimators': [200],
    'max_depth': range(5, 30),
    'learning_rate': scipy.stats.uniform(0.05, 0.15),
    'objective': ['reg:squarederror'],
    'booster': ['gbtree', 'gblinear', 'dart'],
    'min_child_weight': [3, 5, 7],
    'subsample': [0.5, 1, 1.5],
    'colsample_bytree': [0.5, 0.75, 1]
}
    
grid = RandomizedSearchCV(
    XGBRegressor(feature_names=feature_names),
    param_distributions=parameters,
    scoring='neg_mean_absolute_error',  # Use MAE to select the best model
    n_jobs=-1,
    random_state=12345
)  # Use a seed for repeatability
    
start = time()
with joblib.parallel_backend('dask'):
    grid.fit(X, y)

elapsed_time = (time() - start) / 60
print("Training the model took %.2f minutes" % elapsed_time)

以下是我得到的错误：

File "C:\Users\myuser\Documents\code\model.py",
    line 70, in tune_hyperparameters grid.fit(X, y)
File "C:\Users\myuser\Documents\code\condavirtualenv\lib\site-packages\dask_ml\model_selection\_search.py", 
    line 1236, in fit dsk, keys, n_splits, _ = build_cv_graph(
File "C:\Users\myuser\Documents\code\condavirtualenv\lib\site-packages\dask_ml\model_selection\_search.py",
    line 216, in build_cv_graph normalize_estimator(estimator),
File "C:\Users\myuser\Documents\code\condavirtualenv\lib\site-packages\dask_ml\model_selection\_normalize.py",
    line 38, in normalize_estimator val = getattr(est, attr)
File "C:\Users\myuser\Documents\code\condavirtualenv\lib\site-packages\xgboost\sklearn.py",
    line 542, in feature_importances_ b = self.get_booster()
File "C:\Users\myuser\Documents\code\condavirtualenv\lib\site-packages\xgboost\sklearn.py",
     line 193, in get_booster raise XGBoostError('need to call fit or load_model beforehand')
4

0 回答 0