sampler {torch}

Creates a new Sampler

Description

Samplers can be used with dataloader() when creating batches from a torch dataset().

Usage

sampler(
  name = NULL,
  inherit = Sampler,
  ...,
  private = NULL,
  active = NULL,
  parent_env = parent.frame()
)

Arguments

name

(optional) name of the sampler

inherit

(optional) you can inherit from other samplers to re-use some methods.

...

Pass any number of fields or methods. You should at least define the initialize and step methods. See the examples section.

private

(optional) a list of private methods for the sampler

active

(optional) a list of active methods for the sampler.

parent_env

used to capture the right environment to define the class. The default is fine for most situations.

Details

A sampler must implement the .iter and .length() methods.

• initialize takes in a data_source. In general this is a dataset().

• .iter returns a function that returns a dataset index everytime it's called.

• .length returns the maximum number of samples that can be retrieved from that sampler.