py21cmfast.wrapper.arrays.Array¶

class py21cmfast.wrapper.arrays.Array¶

A flexible array management class providing state tracking and initialization capabilities.

The Array class supports dynamic array creation, caching, and state management with immutable semantics.The class allows for creating arrays with configurable shape, data type, initialization function, and optional caching backend. It provides methods for initializing, setting values, removing values, writing to disk, and loading from disk while maintaining a consistent state.

Note

One deliberate exception to this immutability: accessing a purged array’s data directly (e.g. arr.mean(), np.asarray(arr), or any other attribute not defined on this class) - rather than through OutputStruct.get() - transparently loads it from disk and, unless config[“CACHE_ARRAYS_ON_ACCESS”] is set to False, caches the result by mutating _value, state and cache_backend on this instance in place (bypassing frozen=True). Any other reference to the same Array object will observe that mutation. All other methods on this class remain purely functional, returning a new instance rather than mutating self.

Variables:

shape – Dimensions of the array.
dtype – Data type of the array (default is float).
state – Current state of the array.
initfunc – Function used for array initialization (default is np.zeros).
_value – Actual array data, or None if it is not currently in memory (e.g. it has been purged to disk). Private: read the data off the Array itself, or via OutputStruct.get(), so that a purged array is transparently reloaded.
cache_backend – Optional backend for disk caching.

Examples

# Create an array with specific shape and initialize arr = Array(shape=(10, 10)) initialized_arr = arr.initialize()

# Set a value and write to disk arr = arr.with_value(np.random.rand(10, 10)) arr = arr.written_to_disk(backend)

computed() → Self¶

Set the array to a given value and return a new Array.

Return type:: Self

initialize()¶: Initialize the array to its initial/default allocated state.

loaded_from_disk(backend: CacheBackend | None = None) → Self¶

Load values for the array from a cache backend, and return a new instance.

Parameters:: backend (CacheBackend | None)
Return type:: Self

purged_to_disk(backend: CacheBackend | None) → Self¶

Move the array data to disk and return a new object with correct state.

Parameters:: backend (CacheBackend | None)
Return type:: Self

trimmed(trimmed_shape: tuple[int]) → Self¶

Return a new Array with the same data but a different shape, by slicing the original array.

Parameters:: trimmed_shape (tuple[int])
Return type:: Self

with_value(val: numpy.ndarray) → Self¶

Set the array to a given value and return a new Array.

Parameters:: val (numpy.ndarray)
Return type:: Self

without_value() → Self¶

Remove the allocated data from the array.

Return type:: Self

written_to_disk(backend: CacheBackend | None) → Self¶

Write the array to disk and return a new object with correct state.

Parameters:: backend (CacheBackend | None)
Return type:: Self

cache_backend¶

dtype¶

initfunc¶

shape¶

state¶

property value: numpy.ndarray¶

Deprecated accessor for the array’s data.

.value used to be the raw in-memory slot, which meant it read as None for an array that had been purged to disk - a silent wrong answer, since the data was still perfectly available. It now resolves, loading from disk if needed.

It is deprecated because it no longer has a job to do: an OutputStruct’s field attribute (or OutputStruct.get()) already gives you the data as a plain numpy array, so ic.hires_density is what you want. This class is the manager for that data, reached via OutputStruct.arrays[…], and is only of interest when you care about where the data lives rather than what it is.

Return type:: numpy.ndarray