2024 年 8 月 2 日 – ML 函数：改进了 Classification 中的错误消息¶
我们很高兴地宣布，Classification ML 函数的错误消息已得到改进。这些新消息能够更清楚地传达错误的原因，并更好地建议如何解决核心问题。下表显示了最常见的错误消息及其改进版本。
之前的消息 |
新消息 |
|---|---|
Error: Evaluation Data is Not Available.Please train with evaluation enabled. |
Evaluation data is not available.Please create a new model with the EVALUATE parameter set to TRUE.Try adding the following line when calling the model: CONFIG_OBJECT => {'evaluate': "TRUE"}. |
{col} has a different type from training.Make sure each column has the same SQL type between training and prediction. |
Your column {col} has a different SQL type in this dataset than in your training dataset.Make sure each column in this dataset matches the corresponding column's type from your training data.Try casting the column to the type it was in the training data. |
All values in the label column are NULL. |
All values in your target column are NULL.The model requires non-NULL values in your target column to train successfully.Try picking a different target column. |
test_fraction must be a number; test_fraction must be greater than 0 and less than 1 |
Your test_fraction value is not valid. test_fraction must be greater than 0 and less than 1.Try entering a decimal value between 0 and 1 (i.e. 0.2). |
evaluation test_fraction is too large to generate an evaluation dataset |
Your test_fraction is too high.Try a smaller fraction.The test_fraction is likely really close to 1.Try something lower (i.e. 0.2). |
Unable to create an evaluation dataset - only one class present in the evaluation training set.This may be because of a large class imbalance or because of test_fraction is too large. |
Evaluation data is not available.If one class (or category) significantly outweighs the rest, the model may not be able to complete training.Check that each distinct class in your target column appears many times.If the above is not an issue, try lowering your test_fraction. |