InMemoryCacheHolder

class torchdata.datapipes.map.InMemoryCacheHolder(source_dp: MapDataPipe[T_co])

Stores elements from the source DataPipe in memory (functional name: in_memory_cache). Once an item is stored, it will remain unchanged and subsequent retrivals will return the same element. Since items from MapDataPipe are lazily computed, this can be used to store the results from previous MapDataPipe and reduce the number of duplicate computations.

Note

The default cache is a Dict. If another data structure is more suitable as cache for your use

Parameters:

source_dp – source DataPipe from which elements are read and stored in memory

Example

>>> from torchdata.datapipes.map import SequenceWrapper
>>> source_dp = SequenceWrapper(range(10))
>>> cache_dp = source_dp.in_memory_cache()
>>> list(cache_dp)
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]