A tensor backed by a memory-mapped file on disk

Uses R's bigmemory or base file I/O for storage and provides a torch tensor interface for seamless integration. This enables storing tensors larger than available RAM by using disk space.

Value

A DiskTensor object.

Note

R6 doesn't support [ overloading directly; use $get() instead.

Methods

flush(): Flush changes to disk

Methods

`DiskTensor$new()`

Usage

DiskTensor$new(shape, dtype, path, cleanup = TRUE)

Arguments

shape: Integer vector. Shape of the tensor.
dtype: torch.dtype. Data type.
path: Character. File path.
cleanup: Logical. Auto-cleanup on GC.

`DiskTensor$tensor()`

Usage

DiskTensor$tensor()

Returns

A torch tensor.

`DiskTensor$get()`

Usage

DiskTensor$get(indices)

Arguments

indices: List of indices or slices.

Returns

A torch tensor.

`DiskTensor$set()`

Usage

DiskTensor$set(indices, value)

Arguments

indices: List of indices.
value: Tensor. Value to write.

`DiskTensor$flush()`

Usage

DiskTensor$flush()

Returns

Invisible flush to disk

`DiskTensor$nbytes()`

Usage

DiskTensor$nbytes()

Returns

Integer. Size in bytes.

`DiskTensor$clone()`

The objects of this class are cloneable with this method.

Usage

DiskTensor$clone(deep = FALSE)

Arguments

deep: Whether to make a deep clone.

Examples

if (FALSE) { # \dontrun{
# Create a disk-backed tensor
dt <- disk_tensor$new(c(1000L, 128L), torch_float(), "/tmp/my_tensor.mmap")

# Write data (automatically persists to disk)
dt[1:100, ..] <- torch_randn(100L, 128L)

# Read data
x <- dt[1:50, ..]$tensor

# Flush changes to disk
dt$flush()
} # }