机器学习随机森林做特征选择时报编码错误

Question

随机森林做缺失值填充和特征选择时报编码错误，把模型的n_jobs参数删除就能跑，但跑的太慢了，大佬们什么好的解决方法不

特征选择

for i in estimators:
    # 随机森林
    feature_RFC = RandomForestClassifier(n_estimators=i, criterion='gini', n_jobs=-1)
    # 记录分数
    record = []
    # 获得每个特征的重要性
    importance = feature_RFC.fit(feature_train, train_y).feature_importances_
    # 用numpy.linspace创建等差序列
    threshold = np.linspace(0, importance.max(), 10)
    for a in threshold:
        x_embedded = SelectFromModel(feature_RFC, threshold=a).fit_transform(feature_train, train_y)
        score = cross_val_score(feature_RFC, x_embedded, train_y, scoring='roc_auc', cv=5, error_score=np.nan).mean()
        record.append(score)
    # plotly绘图
    fig.add_trace(go.Scatter(x=threshold, y=record, mode='lines', name=i))
fig.show()

报错

  File "E:\Anaconda3\lib\site-packages\joblib\_memmapping_reducer.py", line 590, in register_folder_finalizer
    resource_tracker.register(pool_subfolder, "folder")
  File "E:\Anaconda3\lib\site-packages\joblib\externals\loky\backend\resource_tracker.py", line 191, in register
    self._send('REGISTER', name, rtype)
  File "E:\Anaconda3\lib\site-packages\joblib\externals\loky\backend\resource_tracker.py", line 204, in _send
    msg = '{0}:{1}:{2}\n'.format(cmd, name, rtype).encode('ascii')
UnicodeEncodeError: 'ascii' codec can't encode characters in position 18-21: ordinal not in range(128)

Answer 1

可以考虑修改"E:\Anaconda3\lib\site-packages\joblib\externals\loky\backend\resource_tracker.py"这个本地文件
在204行的_send中找到msg，在其前打印下'{0}:{1}:{2}\n'.format(cmd, name, rtype)，看其中是否存在ascii无法编码的字符，如果存在，删除指定的编码模式"ascii"试试