C++ API Reference for Intel® Data Analytics Acceleration Library 2020 Update 1
Parameter for the Stochastic gradient descent algorithm More...
Parameter | ( | const sum_of_functions::BatchPtr & | function, |
double | momentum = 0.9 , |
||
size_t | nIterations = 100 , |
||
double | accuracyThreshold = 1.0e-05 , |
||
data_management::NumericTablePtr | batchIndices = data_management::NumericTablePtr() , |
||
size_t | batchSize = 128 , |
||
data_management::NumericTablePtr | learningRateSequence = data_management::NumericTablePtr(new data_management::HomogenNumericTable< double >(1, 1, data_management::NumericTableIface::doAllocate, 1.0)) , |
||
size_t | seed = 777 |
||
) |
Constructs the parameter class of the Stochastic gradient descent algorithm
[in] | function | Objective function represented as sum of functions |
[in] | momentum | The momentum value |
[in] | nIterations | Maximal number of iterations of the algorithm |
[in] | accuracyThreshold | Accuracy of the algorithm. The algorithm terminates when this accuracy is achieved |
[in] | batchIndices | Numeric table that represents 32 bit integer indices of terms in the objective function. If no indices are provided, the implementation will generate random indices. |
[in] | batchSize | Number of batch indices to compute the stochastic gradient. If batchSize is equal to the number of terms in objective function then no random sampling is performed, and all terms are used to calculate the gradient. This parameter is ignored if batchIndices is provided. |
[in] | learningRateSequence | Numeric table that contains values of the learning rate sequence |
[in] | seed | Seed for random generation of 32 bit integer indices of terms in the objective function. |
|
virtual |
Checks the correctness of the parameter
Reimplemented from BaseParameter.
double momentum |
Momentum value
For more complete information about compiler optimizations, see our Optimization Notice.