You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Columns of different types are handled as follows:
For numerical columns, the value Object may either be a Double or a String. If a String is passed, then Double.parseDouble() will be called on the String.
For categorical (aka factor, enum) columns, the value Object must be a String with the same names as seen in the training data. It is not allowed to use new categorical (aka factor, enum) levels unseen during training (this will result in a hex.genmodel.easy.exception.PredictUnknownCategoricalLevelException when one of the predict methods is called).
We can update the wrappers to create Doubles or Strings depending on what is in the NDArray ListValue, but the above docs suggest you should send categorical data as Strings.
https://github.com/SeldonIO/seldon-core/blob/master/wrappers/s2i/java/wrapper/src/main/java/io/seldon/wrapper/utils/H2OUtils.java#L71
We should check the type of the items and convert to appropriate type if possible.
The text was updated successfully, but these errors were encountered: