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:null
.label
.index
.prob
.\"0\"
, \"1\"
, etc. Default value is null
.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.true
, also returns a List<List<Double>>
of all probabilities (basically, convert NDArray to list). false
by default.