dense

Type Members

1. class AdadeltaGradientDescentDVD extends StochasticGradientDescent[DenseVector[Double]]

   

2. case class AffineOutputTransform[FV](numOutputs: Int, numInputs: Int, innerTransform: Transform[FV, DenseVector[Double]], includeBias: Boolean = true) extends OutputTransform[FV, DenseVector[Double]] with Product with Serializable

   

   Used at the output layer when we're only going to need some of the possible ouputs; it exposes the penultimate layer and then the Layer allows you to pass the results from that back in (caching it elsewhere) and only compute certain cells in the output layer (activationsFromPenultimateDot).

3. case class AffineTransform[FV, Mid](numOutputs: Int, numInputs: Int, innerTransform: Transform[FV, Mid], includeBias: Boolean = true)(implicit mult: breeze.linalg.operators.OpMulMatrix.Impl2[DenseMatrix[Double], Mid, DenseVector[Double]], canaxpy: breeze.linalg.scaleAdd.InPlaceImpl3[DenseVector[Double], Double, Mid]) extends Transform[FV, DenseVector[Double]] with Product with Serializable

   

4. case class BatchNormalizationTransform[FV](size: Int, useBias: Boolean, inner: Transform[FV, DenseVector[Double]]) extends Transform[FV, DenseVector[Double]] with Product with Serializable

   

   Implements batch normalization from http://arxiv.org/pdf/1502.03167v3.pdf Basically, each unit is shifted and rescaled per minibatch so that its activations have mean 0 and variance 1.

   Implements batch normalization from http://arxiv.org/pdf/1502.03167v3.pdf Basically, each unit is shifted and rescaled per minibatch so that its activations have mean 0 and variance 1. This has been demonstrated to help training deep networks, but doesn't seem to help here.

5. case class CachingLookupAndAffineTransformDense[FV](numOutputs: Int, numInputs: Int, word2vecIndexed: Word2VecIndexed[String], includeBias: Boolean = true) extends Transform[Array[Int], DenseVector[Double]] with Product with Serializable

   

   Used at the input layer to cache lookups and the result of applying the affine transform at the first layer of the network.

   Used at the input layer to cache lookups and the result of applying the affine transform at the first layer of the network. This saves computation across repeated invocations of the neural network in the sentence.

6. case class CachingLookupTransform(word2vecIndexed: Word2VecIndexed[String]) extends Transform[Array[Int], DenseVector[Double]] with Product with Serializable

   

   Used at the input layer to cache lookups and

7. case class EmbeddingsTransform[FV](numOutputs: Int, numInputs: Int, word2vecIndexed: Word2VecIndexed[String], includeBias: Boolean = true) extends Transform[Array[Int], DenseVector[Double]] with Product with Serializable

   

   Used at the input layer to cache lookups and backprop into embeddings

8. class FrequencyTagger[W] extends Tagger[W] with Serializable

   

9. class IdentityTransform[T] extends Transform[T, T]

   

10. case class LRQTNLayer(lhsWeights: DenseMatrix[Double], rhsWeights: DenseMatrix[Double], index: Index[Feature], numRanks: Int, numLeftInputs: Int, numRightInputs: Int) extends Product with Serializable

    

11. case class LowRankQuadraticTransform[FV](numOutputs: Int, numRanks: Int, numLeftInputs: Int, numRightInputs: Int, innerTransform: Transform[FV, DenseVector[Double]]) extends OutputTransform[FV, DenseVector[Double]] with Product with Serializable

    

12. case class LowRankQuadraticTransformNeuron(numRanks: Int, numLeftInputs: Int, numRightInputs: Int) extends Product with Serializable

    

    Separate because I was having some issues...

13. case class NeuralBias(input: Int) extends Feature with Product with Serializable

    

14. case class NeuralFeature(output: Int, input: Int) extends Feature with Product with Serializable

    

15. case class NonlinearTransform[FV](nonLinType: String, size: Int, inner: Transform[FV, DenseVector[Double]], dropoutRate: Double = 0.5) extends Transform[FV, DenseVector[Double]] with Product with Serializable

    

    A bit of a misnomer since this has been generalized to support linear functions as well...

16. case class OutputEmbeddingTransform[FV](numOutputs: Int, outputDim: Int, innerTransform: Transform[FV, DenseVector[Double]], coarsenerForInitialization: Option[(Int) ⇒ Int] = None) extends OutputTransform[FV, DenseVector[Double]] with Product with Serializable

    

    Output embedding technique described in section 6 of http://www.eecs.berkeley.edu/~gdurrett/papers/durrett-klein-acl2015.pdf Basically learns a dictionary for the output as well as an affine transformation in order to produce the vector that gets combined with the input in the final bilinear product.

17. trait OutputTransform[In, +Out] extends Serializable

    

18. trait Tagger[W] extends AnyRef

    

19. case class TanhTransform[FV](inner: Transform[FV, DenseVector[Double]]) extends Transform[FV, DenseVector[Double]] with Product with Serializable

    

20. trait Transform[In, +Out] extends Serializable

    

21. class Word2VecDepFeaturizerIndexed[W] extends Serializable

    

22. class Word2VecIndexed[W] extends Serializable

    

    converter is used to map words into the word2vec vocabulary, which might include things like lowercasing, replacing numbers, changing -LRB-, etc.

    converter is used to map words into the word2vec vocabulary, which might include things like lowercasing, replacing numbers, changing -LRB-, etc. See Word2Vec.convertWord

23. class Word2VecSurfaceFeaturizerIndexed[W] extends Serializable

    

24. class Word2VecUtils extends AnyRef

    

25. trait WordVectorAnchoringIndexed[String] extends AnyRef

    

26. trait WordVectorDepAnchoringIndexed[String] extends AnyRef