LF_ApplyClassification Processing Resource

The classification application PR allows you to apply a classifier you already learned using the classification training PR. The PR has no init time parameters. Here are the runtime parameters:

• algorithmParameters Parameters to pass on to the application algorithm, if any.
• confidenceThreshold (double, default: missing, do not use) - the minimum confidence score threshold required for a classification to get assigned (this has no effect if the algorithm does not produce a classification confidence score). If the minimum confidence is not reached, the instance annotation is not created or does not get the targetFeature set or overriden. If this parameter is left unspecified, no checking for a minimum confidenceThreshold is performed.
• dataDirectory The directory that was used to save the model during training.
• inputASName The annotation set that contains the instance annotations and the annotations specified in the feature configuration file.
• instanceType The annotation type of the instance annotations.
• outputASName The annotation set where the prediction will be placed. If this is the same as the input annotation set, then the existing instance annotations will be updated, otherwise new annotation will be created.
• sequenceSpan The annotation type for sequence spans if the classification algorithm is a sequence learning algorithm
• serverUrl (String, no default) if specified, will try to connect to the given URL and use the server there to get the predictions. See ServerForApplication for details.
• targetFeature Which feature to write the classification onto. Leave blank to put it in the feature that was learned at training time.

AlgorithmParameters

If a serverUrl is specified and thus a HTTP server is used to carry out the classification, the following parameters are supported:

• -d or -dense: send the vectors in dense format, default is sparse format