FAQ（常见问题解答）

Can I use objects or media sets as inputs in Model Studio?

No, the only inputs accepted in Model Studio are datasets. To achieve a similar result to using an object input, you can use the object's backing datasource as an input.

Why are my modeling objective evaluation results showing my model as 100% accurate?

The default binary classification and regression evaluators operate under the assumption that the output dataset will contain both the target and predicted column when running inference over the input dataset. However, Model Studio trainers default to setting the prediction column name to the target column name, which will overwrite the target column during the evaluation inference job and produce a dataset that does not adhere to the evaluator's assumption.

To fix this, you can override the prediction column name in the model studio configuration. Manually set a prediction column name that is different from the target column name so the model does not overwrite it.

---

中文翻译

常见问题解答

我能否在Model Studio中使用对象(Object)或媒体集(Media Set)作为输入？

不可以，Model Studio仅接受数据集(Dataset)作为输入。如需实现与使用对象输入类似的效果，您可以使用该对象的底层数据源(Backing Datasource)作为输入。

为什么我的建模目标评估结果显示模型准确率为100%？

默认的二元分类和回归评估器(Evaluator)在运行推理时，会假设输出数据集同时包含目标列(Target Column)和预测列(Predicted Column)。然而，Model Studio的训练器(Trainer)默认将预测列名称设置为与目标列名称相同，这会导致在评估推理作业中覆盖目标列，从而生成不符合评估器假设的数据集。

要解决此问题，您可以在Model Studio配置中覆盖预测列名称。手动设置一个与目标列名称不同的预测列名称，这样模型就不会覆盖目标列。