[Catboost][ClearML]错误:如果损失函数是对数损失，那么对于0和1类，应该给出类的权重。

Question

在最近开始使用ClearML来管理MLOps之后，我面临着以下问题:当运行一个脚本，使用计算机中不同的类权重在二进制分类问题中训练CatBoost时，它完美地工作，记录结果，完全没有问题。一旦我尝试使用ClearML代理远程运行该程序，它将导致以下错误：

<!-- language: lang-none -->
Traceback (most recent call last):
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/clearml/binding/frameworks/catboost_bind.py", line 102, in _fit
    return original_fn(obj, *args, **kwargs)
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/catboost/core.py", line 5007, in fit
    self._fit(X, y, cat_features, text_features, embedding_features, None, sample_weight, None, None, None, None, baseline, use_best_model,
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/catboost/core.py", line 2262, in _fit
    train_params = self._prepare_train_params(
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/catboost/core.py", line 2194, in _prepare_train_params
    _check_train_params(params)
  File "_catboost.pyx", line 6032, in _catboost._check_train_params
  File "_catboost.pyx", line 6051, in _catboost._check_train_params
**_catboost.CatBoostError: catboost/private/libs/options/catboost_options.cpp:607: if loss-function is Logloss, then class weights should be given for 0 and 1 classes
During handling of the above exception, another exception occurred:
Traceback (most recent call last):**
  File "/root/.clearml/venvs-builds/3.9/task_repository/RecSys.git/src/cli/model_training_remote.py", line 313, in <module>
    rfs.run(
  File "/root/.clearml/venvs-builds/3.9/task_repository/RecSys.git/src/cli/model_training_remote.py", line 232, in run
    model.fit(
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/clearml/binding/frameworks/__init__.py", line 36, in _inner_patch
    raise ex
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/clearml/binding/frameworks/__init__.py", line 34, in _inner_patch
    ret = patched_fn(original_fn, *args, **kwargs)
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/clearml/binding/frameworks/catboost_bind.py", line 110, in _fit
    return original_fn(obj, *args, **kwargs)
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/catboost/core.py", line 5007, in fit
    self._fit(X, y, cat_features, text_features, embedding_features, None, sample_weight, None, None, None, None, baseline, use_best_model,
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/catboost/core.py", line 2262, in _fit
    train_params = self._prepare_train_params(
  File "/root/.clearml/venvs-builds/3.9/lib/python3.9/site-packages/catboost/core.py", line 2194, in _prepare_train_params
    _check_train_params(params)
  File "_catboost.pyx", line 6032, in _catboost._check_train_params
  File "_catboost.pyx", line 6051, in _catboost._check_train_params
**_catboost.CatBoostError: catboost/private/libs/options/catboost_options.cpp:607: if loss-function is Logloss, then class weights should be given for 0 and 1 classes**

我确实把字典连接起来了：

    model_params = {
        "loss_function": "Logloss",
        "eval_metric": "AUC",
        "class_weights": {0: 1, 1: 60},
        "learning_rate": 0.1
    }

在ClearML任务中注册为

task.connect(model_params, 'model_params')

并在以下调用中用作模型的参数：

model = CatBoostClassifier(**model_params)

在ClearML交互模式下从容器中运行它时，它也工作得很好。

Answer 1

免责声明:我是ClearML的团队成员

我想我理解这个问题，基本上我认为问题是：

task.connect(model_params, 'model_params')

因为这是一个嵌套的dict：

    model_params = {
        "loss_function": "Logloss",
        "eval_metric": "AUC",
        "class_weights": {0: 1, 1: 60},
        "learning_rate": 0.1
    }

class_weights存储为String密钥，但catboost需要int密钥，因此失败。一种选择是删除task.connect(model_params, 'model_params')

另一个解决方案(直到我们修复它)将是这样做：

task.connect(model_params, 'model_params')
model_params["class_weights"] = {
0: model_params["class_weights"].get("0", model_params["class_weights"].get(0))
1: model_params["class_weights"].get("1", model_params["class_weights"].get(1))
}