Bayesian Optimization代码报错：无法从colorama导入just_fix_windows_console

问题详情 #

使用Anaconda构建机器学习模型，运行贝叶斯优化代码时持续报错：ImportError: cannot import name 'just_fix_windows_console' from 'colorama'，已安装并更新所有相关包，但问题仍未解决，需要让代码正常执行贝叶斯优化。

执行代码 #

from bayes_opt import BayesianOptimization

def fcv(max_depth, gamma, min_child_weight, subsample, colsample_bytree, learning_rate, num_boost_round):
  params = {"objective":'reg:squarederror', "max_depth":int(max_depth), 'gamma':gamma, 'min_child_weight':min_child_weight, 'subsample':subsample, "colsample_bytree":colsample_bytree, "learning_rate":learning_rate}
  cv_results=xgb.cv(dtrain=data_dmatrix, params=params, nfold=10, num_boost_round=int(num_boost_round), early_stopping_rounds=10, metrics='rmse', as_pandas=True)
  return -cv_results['test-rmse-mean'].min() 


dict_cv = {
          'max_depth': (2, 12),
          'gamma': (0.001, 10.0),
          'min_child_weight': (0, 20),
          'subsample': (0.4, 1.0),
          'colsample_bytree': (0.4, 1.0),
          'learning_rate': (0.1, 1.0),
          'num_boost_round' :(30, 100)
          }

XGB_BO = BayesianOptimization(fcv, dict_cv) #Creating the optimizer
XGB_BO.maximize(init_points=10, n_iter=40, acq='ei', xi=0.0) #Running optimization!

报错信息 #

Output exceeds the size limit. Open the full output data in a text editor
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
c:\Users\ehler\Documents\Masterarbeit\Codes\MAcoding_v5.ipynb Zelle 34 in <cell line: 1>()
----> 1 from bayes_opt import BayesianOptimization
      2 # Gradient Boosting Machine
      3 def gbm_cl_bo(max_depth, max_features, learning_rate, n_estimators, subsample):

File c:\Users\ehler\anaconda3\lib\site-packages\bayes_opt\__init__.py:1, in <module>
----> 1 from .bayesian_optimization import BayesianOptimization, Events
      2 from .domain_reduction import SequentialDomainReductionTransformer
      3 from .util import UtilityFunction

File c:\Users\ehler\anaconda3\lib\site-packages\bayes_opt\bayesian_optimization.py:5, in <module>
      1 import warnings
      3 from bayes_opt.constraint import ConstraintModel
----> 5 from .target_space import TargetSpace
      6 from .event import Events, DEFAULT_EVENTS
      7 from .logger import _get_default_logger

File c:\Users\ehler\anaconda3\lib\site-packages\bayes_opt\target_space.py:4, in <module>
      1 import warnings
      3 import numpy as np
----> 4 from .util import ensure_rng, NotUniqueError
      5 from .util import Colours
      8 def _hashable(x):
...
   (...)
     44     :return: x_max, The arg max of the acquisition function.
     45     """

ImportError: cannot import name 'just_fix_windows_console' from 'colorama' (c:\Users\ehler\anaconda3\lib\site-packages\colorama\__init__.py)

解决方案 #

• 升级colorama版本：just_fix_windows_console是colorama 0.4.6及以上版本新增的函数，低版本没有该接口。

  1. 查看当前colorama版本：

     conda list colorama
     

  2. 如果版本低于0.4.6，执行升级：

     conda install -c conda-forge colorama>=0.4.6
     

• 重新安装bayesian-optimization：可能是包依赖不匹配导致的问题，卸载后重装可解决版本冲突：

  pip uninstall bayesian-optimization -y
  pip install bayesian-optimization
  

  或者使用conda安装：

  conda install -c conda-forge bayesian-optimization
  

• 手动修改bayes_opt源码（临时方案）：如果上述方法无效，可直接跳过控制台修复逻辑：
  找到c:\Users\ehler\anaconda3\lib\site-packages\bayes_opt\util.py文件，找到以下代码并注释掉：

  # from colorama import just_fix_windows_console
  # just_fix_windows_console()
  

  该操作仅跳过控制台颜色修复，不会影响贝叶斯优化的核心功能。

内容的提问来源于stack exchange，提问作者Christoph Ehler