Packages

package cntk

Ordering
  1. Alphabetic
Visibility
  1. Public
  2. All

Type Members

  1. class CNTKFunctionParam extends ComplexParam[SerializableFunction] with ParamEquality[SerializableFunction]

    Param for ByteArray.

    Param for ByteArray. Needed as spark has explicit params for many different types but not ByteArray.

  2. class CNTKModel extends Model[CNTKModel] with ComplexParamsWritable with HasMiniBatcher with HasFeedFetchDicts with Wrappable with BasicLogging
  3. class ImageFeaturizer extends Transformer with HasInputCol with HasOutputCol with Wrappable with ComplexParamsWritable with BasicLogging

    The ImageFeaturizer relies on a CNTK model to do the featurization, one can set this model using the modelLocation parameter.

    The ImageFeaturizer relies on a CNTK model to do the featurization, one can set this model using the modelLocation parameter. To map the nodes of the CNTK model onto the standard "layers" structure of a feed forward neural net, one needs to supply a list of node names that range from the output node, back towards the input node of the CNTK Function. This list does not need to be exhaustive, and is provided to you if you use a model downloaded from the ModelDownloader, one can find this layer list in the schema of the downloaded model.

    The ImageFeaturizer takes an input column of images (the type returned by the ImageReader), and automatically resizes them to fit the CMTKModel's inputs. It then feeds them through a pre-trained CNTK model. One can truncate the model using the cutOutputLayers parameter that determines how many layers to truncate from the output of the network. For example, layer=0 means that no layers are removed, layer=2 means that the image featurizer returns the activations of the layer that is two layers from the output layer.

Ungrouped