CLASSIFIER_OUTPUT

ClassifierOutputStep takes as input a numerical 2d NDArray (i.e., float/double etc. type) with shape [minibatch, numClasses] which represents the softmax predictions for a standard classifier and returns based on this array:
The predicted class label - as a String
The predicted class index - as an integer (long)
The predicted class probability - as a Double
Configs
Descriptions
inputNames
Optional. If set: this represents the NDArray. If not set: use DataUtils.inferField to find an NDArray field. Default is null.
returnLabel
Default is true; if false, don't return label.
returnIndex
Default is true; if false, don't return index.
returnProb
Default is true; if false, don't return probability.
labelName
Output names for the labels. Default value is label.
indexName
Output names for the index. Default value is index.
probName
Output names for the labels probabilities. Default value is prob.
labels
As a List<String>. Optional. If not specified, the predicted class index as a string is used - i.e., \"0\", \"1\", etc. Default value is null.
topN
Integer, null by default. If non-null and > 1, we return List<String>, List<Long>, List<Double> for the predicted class/index/probability instead of String/Long/Double.
allProbabilities
If true, also returns a List<List<Double>> of all probabilities (basically, convert NDArray to list). false by default.

SAMEDIFF

REGRESSION_OUTPUT

Last updated 3 months ago

Configs	Descriptions
inputNames	Optional. If set: this represents the NDArray. If not set: use DataUtils.inferField to find an NDArray field. Default is `null`.
returnLabel	Default is true; if false, don't return label.
returnIndex	Default is true; if false, don't return index.
returnProb	Default is true; if false, don't return probability.
labelName	Output names for the labels. Default value is `label`.
indexName	Output names for the index. Default value is `index`.
probName	Output names for the labels probabilities. Default value is `prob`.
labels	As a List<String>. Optional. If not specified, the predicted class index as a string is used - i.e., `\"0\"`, `\"1\"`, etc. Default value is `null`.
topN	Integer, `null` by default. If non-null and > 1, we return `List<String>`, `List<Long>`, `List<Double>` for the predicted class/index/probability instead of String/Long/Double.
allProbabilities	If `true`, also returns a `List<List<Double>>` of all probabilities (basically, convert NDArray to list). `false` by default.