Deep.Net


Dataset

Namespace: Datasets

Dataset functions.

Functions and values

Function or valueDescription
batches batchSize ds
Signature: batchSize:int64 -> ds:Dataset<'S> -> seq<'S>
Type parameters: 'S

Returns a sequence of batches with the given size of this dataset.

cutSequences stepsPerCut ds
Signature: stepsPerCut:int64 -> ds:Dataset<'?175694> -> Dataset<'?175695>
Type parameters: '?175694, '?175695

Cuts each sequence in the dataset into multiple chunks of length stepsPerCut.

cutToMinSamples minSamples ds
Signature: minSamples:int64 -> ds:Dataset<'?175697> -> Dataset<'?175697>
Type parameters: '?175697

Cuts each sequence in the dataset into multiple chunks so that at least minSamples are in the resulting dataset. If the dataset already contains at least minSamples, then the function returns the dataset unaltered.

map f ds
Signature: f:(ITensor -> '?175699) -> ds:Dataset<'S> -> Dataset<'S>
Type parameters: '?175699, 'S

maps the field storages using the given function creating a new dataset

nSamples ds
Signature: ds:Dataset<'?175690> -> int64
Type parameters: '?175690

Number of samples.

nSteps ds
Signature: ds:Dataset<'?175692> -> int64
Type parameters: '?175692

Number of steps per sample in a sequence dataset.

ofSamples samples
Signature: samples:seq<'?175680> -> Dataset<'?175680>
Type parameters: '?175680

Constructs a dataset from a sequence of samples of record type 'S. Each field in 'S must be of type ArrayNDT<_> and the dimensionality of each field must be constant over all samples. If the shape of a field varies over the samples it is padded (with zeros) to the largest shape in the sample sequence. The given sequence is enumerated only one time and the data is copied once.

ofSeqSamples seqSamples
Signature: seqSamples:seq<'?175682> -> Dataset<'?175682>
Type parameters: '?175682

Constructs a sequence dataset from a sequence of samples of record type 'S.

partition ratios ds
Signature: ratios:float<MeasureOne> list -> ds:Dataset<'S> -> Dataset<'S> list
Type parameters: 'S

Partitions this dataset using the given ratios.

slotBatches batchSize slotSize ds
Signature: batchSize:int64 -> slotSize:int64 -> ds:Dataset<'S> -> seq<'S>
Type parameters: 'S

Returns a sequence of batches of time slots with size batchSize and slotSize respectively of the dataset.

toCuda ds
Signature: ds:Dataset<'S> -> Dataset<'S>
Type parameters: 'S

copies this dataset to a CUDA GPU

toHost ds
Signature: ds:Dataset<'S> -> Dataset<'S>
Type parameters: 'S

copies this dataset to the host

transfer dev ds
Signature: dev:ITensorDevice -> ds:Dataset<'S> -> Dataset<'S>
Type parameters: 'S

copies this dataset to the specified device

Fork me on GitHub