Type Members

1.  abstract class AbstractOptimizer extends AnyRef
2.  class AccuracyResult extends ValidationResult

   Represent an accuracy result.

   Represent an accuracy result. Accuracy means a ratio of correct number and total number.

3.  class Adadelta[T] extends OptimMethod[T]

   Adadelta implementation for SGD: http://arxiv.org/abs/1212.5701

4.  class Adagrad[T] extends OptimMethod[T]

   An implementation of Adagrad.

   An implementation of Adagrad. See the original paper: http://jmlr.org/papers/volume12/duchi11a/duchi11a.pdf

5.  class Adam[T] extends OptimMethod[T]

   An implementation of Adam http://arxiv.org/pdf/1412.6980.pdf

6.  class Adamax[T] extends OptimMethod[T]

   An implementation of Adamax http://arxiv.org/pdf/1412.6980.pdf

7.  class ArrayBufferAccumulator extends AccumulatorV2[ArrayBuffer[Double], ArrayBuffer[Double]]
8.  class ContiguousResult extends ValidationResult

   A generic result type who's data is contiguous float.

9.  class DistriOptimizer[T] extends Optimizer[T, MiniBatch[T]]

   The optimizer run on a distributed cluster.

10.  class DistriOptimizerV2[T] extends Optimizer[T, MiniBatch[T]]

    The optimizer run on a distributed cluster.

11.  class DistriValidator[T] extends Validator[T, MiniBatch[T]]

    Validate model on a distributed cluster.

12.  class Evaluator[T] extends Serializable

    model evaluator

13.  class Ftrl[T] extends OptimMethod[T]

    An implementation of Ftrl https://www.eecs.tufts.edu/~dsculley/papers/ad-click-prediction.pdf.

    An implementation of Ftrl https://www.eecs.tufts.edu/~dsculley/papers/ad-click-prediction.pdf. Support L1 penalty, L2 penalty and shrinkage-type L2 penalty.

14.  class HitRatio[T] extends ValidationMethod[T]

    Hit Ratio(HR).

    Hit Ratio(HR). HR intuitively measures whether the test item is present on the top-k list.

15.  class L1L2Regularizer[T] extends Regularizer[T]

    Apply both L1 and L2 regularization

    Apply both L1 and L2 regularization

    T

    type parameters Float or Double

    Annotations

    @SerialVersionUID()

16.  case class L1Regularizer[T](l1: Double)(implicit evidence$3: ClassTag[T], ev: TensorNumeric[T]) extends L1L2Regularizer[T] with Product with Serializable

    Apply L1 regularization

    Apply L1 regularization

    T

    type parameters Float or Double

    l1

    l1 regularization rate

    Annotations

    @SerialVersionUID()

17.  case class L2Regularizer[T](l2: Double)(implicit evidence$4: ClassTag[T], ev: TensorNumeric[T]) extends L1L2Regularizer[T] with Product with Serializable

    Apply L2 regularization

    Apply L2 regularization

    T

    type parameters Float or Double

    l2

    l2 regularization rate

    Annotations

    @SerialVersionUID()

18.  class LBFGS[T] extends OptimMethod[T]

    This implementation of L-BFGS relies on a user-provided line search function (state.lineSearch).

    This implementation of L-BFGS relies on a user-provided line search function (state.lineSearch). If this function is not provided, then a simple learningRate is used to produce fixed size steps. Fixed size steps are much less costly than line searches, and can be useful for stochastic problems.

    The learning rate is used even when a line search is provided. This is also useful for large-scale stochastic problems, where opfunc is a noisy approximation of f(x). In that case, the learning rate allows a reduction of confidence in the step size.

19.  class LarsSGD[T] extends SGD[T]

    An implementation of LARS https://arxiv.org/abs/1708.03888 Lars.createOptimForModule is recommended to be used to create LARS optim methods for multiple layers

20.  trait LineSearch[T] extends AnyRef

    Line Search strategy

21.  class LocalOptimizer[T] extends Optimizer[T, MiniBatch[T]]

    Optimize a model on a single machine

22.  class LocalPredictor[T] extends Serializable

    Predictor for local data

23.  class LocalValidator[T] extends Validator[T, MiniBatch[T]]

    Validate a model on a single machine Use given dataset with certain validation methods such as Top1Accuracy as an argument of its test method

24.  class Loss[T] extends ValidationMethod[T]

    This evaluation method is calculate loss of output with respect to target

25.  class LossResult extends ContiguousResult

    Use loss as a validation result

26.  case class LossWithElapsedTime(index: Int, loss: Double, elapsed: Long) extends Product with Serializable
27.  class MAE[T] extends ValidationMethod[T]

    This evaluation method is calculate mean absolute error of output with respect to target

28.  class MAPMultiIOUValidationResult extends ValidationResult
29.  class MAPType extends Serializable
30.  class MAPValidationResult extends ValidationResult

    The MAP Validation Result.

    The MAP Validation Result. The results are not calculated until result() or format() is called require class label beginning with 0

31.  class MeanAveragePrecision[T] extends ValidationMethod[T]

    Calculate the Mean Average Precision (MAP).

    Calculate the Mean Average Precision (MAP). The algorithm follows VOC Challenge after 2007 Require class label beginning with 0

32.  class MeanAveragePrecisionObjectDetection[T] extends ValidationMethod[T]

    MeanAveragePrecision for Object Detection The class label begins with 0

    MeanAveragePrecision for Object Detection The class label begins with 0

    The expected output from the last layer should be a Tensor[Float] or a Table If output is a tensor, it should be [num_of_batch X (1 + maxDetection * 6)] matrix The format of the matrix should be [<batch>, <batch>, ...], where each row vector is <batch> = [<size_of_batch>, <sample>,...]. Each sample has format: <sample> = <label, score, bbox x4> imgId is the batch number of the sample. imgId begins with 0. Multiple samples may share one imgId

    If output is a table, it is a table of tables. output(i) is the results of the i-th image in the batch, where i = 1 to sizeof(batch) output(i) is a table, which contains the same keys (fields) of image info in the "target" Please refer to RoiMiniBatch/RoiImageInfo's documents. Besides, the inner tables also contain the scores for the detections in the image.

    The "target" (Ground truth) is a table with the same structure of "output", except that it does not have "score" field

33.  class Metrics extends Serializable

    Performance metrics for the training process.

    Performance metrics for the training process. Beyond collect local metrics(e.g. throughput) in driver node, it can also be used to collect distributed metrics (e.g. time of some steps among the workers). The is useful for performance analysis.

34.  class NDCG[T] extends ValidationMethod[T]

    Normalized Discounted Cumulative Gain(NDCG).

    Normalized Discounted Cumulative Gain(NDCG). NDCG accounts for the position of the hit by assigning higher scores to hits at top ranks.

35.  trait OptimMethod[T] extends Serializable

    Similar to torch Optim method, which is used to update the parameter

36.  abstract class Optimizer[T, D] extends AnyRef

    Optimizer is an abstract class which is used to train a model automatically with some certain optimization algorithms.

    Optimizer is an abstract class which is used to train a model automatically with some certain optimization algorithms.

    T

    numeric type, which can be Float or Double

    D

    the type of elements in DataSet, such as MiniBatch

37.  trait OptimizerLogger extends AnyRef
38.  class PRAUCResult extends ValidationResult
39.  class ParallelAdam[T] extends OptimMethod[T]

    An multi-thread implementation of Adam http://arxiv.org/pdf/1412.6980.pdf

40.  class ParallelOptimizer[T] extends Optimizer[T, MiniBatch[T]]

    The optimizer run on a distributed cluster.

41.  case class ParamSegments[T](start: Int, length: Int, method: OptimMethod[T]) extends Product with Serializable
42.  class PrecisionRecallAUC[T] extends ValidationMethod[T]

    Precision Recall Area Under Curve will compute the precision-recall pairs and get the area under the curve.

    Precision Recall Area Under Curve will compute the precision-recall pairs and get the area under the curve.

    Note: It will gather all output probabilities and targets to driver and will compute the precision, recall and the auc every calling of result()

    T

    class tag for tensor numeric

43.  class PredictionService[T] extends AnyRef

    In this service, concurrency is kept not greater than numThreads by a BlockingQueue, which contains available model instances.

    Thread-safe Prediction Service for Concurrent Calls

    In this service, concurrency is kept not greater than numThreads by a BlockingQueue, which contains available model instances.
    
    numThreads model instances sharing weights/bias will be put into the BlockingQueue during initialization.
    
    When predict method called, service will try to take an instance from BlockingQueue, which means if all instances are on serving, the predicting request will be blocked until some instances are released.
    
    If exceptions caught during predict, a scalar Tensor[String] will be returned with thrown message.

44.  class Predictor[T] extends Serializable

    Predictor for distributed data

    Predictor for distributed data

    NOTE: The predictClass, predict and predictImage will call the relevant methods of object Predictor. Why we do this? Because every these methods uses the ClassTag T. If we do these jobs in the methods of classPredictor, when we do mapPartition, Spark will find all used values and do serialization. The T is the argument of constructor, the serialization will package the whole Predictor class, which contains themodel. It will send a duplicate model to the workers. So we should move these methods to object Predictor.

45.  class RMSprop[T] extends OptimMethod[T]

    An implementation of RMSprop

46.  trait Regularizer[T] extends Serializable

    It is a trait for all regularizers.

    It is a trait for all regularizers. Any regularizers need to inherit the result.

    T

    type parameters Float or Double

47.  class SGD[T] extends OptimMethod[T]

    A plain implementation of SGD

48.  class Top1Accuracy[T] extends ValidationMethod[T]

    Caculate the percentage that output's max probability index equals target

49.  class Top5Accuracy[T] extends ValidationMethod[T]

    Calculate the percentage that target in output's top5 probability indexes

50.  class TrainingContext[T] extends Serializable
51.  class TreeNNAccuracy[T] extends ValidationMethod[T]

    This is a metric to measure the accuracy of Tree Neural Network/Recursive Neural Network

52.  trait Trigger extends Serializable

    A trigger specifies a timespot or several timespots during training, and a corresponding action will be taken when the timespot(s) is reached.

53.  trait ValidationMethod[T] extends Serializable

    A method defined to evaluate the model.

    A method defined to evaluate the model. This trait can be extended by user-defined method. Such as Top1Accuracy

54.  trait ValidationResult extends Serializable

    A result that calculate the numeric value of a validation method.

    A result that calculate the numeric value of a validation method. User-defined valuation results must override the + operation and result() method. It is executed over the samples in each batch.

55.  abstract class Validator[T, D] extends AnyRef

    Validator is an abstract class which is used to test a model automatically with some certain validation methods such as Top1Accuracy, as an argument of its test method.

    Validator is an abstract class which is used to test a model automatically with some certain validation methods such as Top1Accuracy, as an argument of its test method.

    T

    numeric type, which can be Float or Double

    D

    the type of elements in DataSet, such as MiniBatch