fosanalysis.utils.windows Namespace Reference

Contains functions to shift a (partly) view of an array across the full array. More...

Functions
list	determine_moving_parameters (np.array data_array, tuple radius, tuple start_pixel=None, tuple step_size=None)
	Generate indices for a moving window and check the other parameters.
	moving (np.array data_array, tuple radius, tuple start_pixel=None, tuple step_size=None)
	Generates a symmetric moving window over an array.
	sliding (np.array data_array, radius)
	Generates a sliding window over an array.
np.array	sliding_window_function (np.array arr, radius, fn, str pad_mode=None, *args, **kwargs)
	Applies the function fn to a sliding window over the array arr.

Detailed Description

Contains functions to shift a (partly) view of an array across the full array.

Author

Bertram Richter

Date

2024

Function Documentation

determine_moving_parameters()

list fosanalysis.utils.windows.determine_moving_parameters	(	np.array	data_array,
		tuple	radius,
		tuple	start_pixel = None,
		tuple	step_size = None )

Generate indices for a moving window and check the other parameters.

Parameters

data_array	Array of data over which the window should move.
radius	Inradius of the window's rectangle. If radius is an int, all axes will use this radius and the window is a square. For non-square windows, pass a tuple with a radius for each dimension of data_array. Along an axis, the window has a width of \(2r + 1\) for each element \(r\) of radius.
start_pixel	Index of the first window's central pixel. If start_pixel is an int, it is used for all dimensions of data_array. To specify a custom starting element, pass a tuple with a step size for each dimension of data_array. If None, it defaults to radius, the moving window starts with a full slice.
step_size	Step size for estimation. If step_size is an int, it is used for all dimensions of data_array. If None, it defaults to \(2r + 1\) for each element \(r\) of radius, which is equivalent to a rolling window.

Returns

Retuns a tuple like (orig_index_lists, radius, start_pixel, step_size):

Return values

orig_index_lists	List of lists, containing the indices along each axis.
radius	A tuple, see above.
start_pixel	A tuple, see above.
step_size	A tuple, see above.

Definition at line 158 of file windows.py.

Here is the caller graph for this function:

moving()

fosanalysis.utils.windows.moving	(	np.array	data_array,
		tuple	radius,
		tuple	start_pixel = None,
		tuple	step_size = None )

Generates a symmetric moving window over an array.

This function returns a generator that yields information about the window_center_orig, target_pixel, and the content of the window. Converts data_array to a Numpy array. Ensuring radius as a tuple. do this similar for stepsize and the start pixel (start pixel if not given equal to radius). Determine orig_index_lists based on array dimensions and provided indices. It generates combinations of indices. It iterates over combinations and yields window information.

Parameters

data_array	Array of data over which the window should move.
radius	Inradius of the window's rectangle. If radius is an int, all axes will use this radius and the window is a square. For non-square windows, pass a tuple with a radius for each dimension of data_array. Along an axis, the window has a width of \(2r + 1\) for each element \(r\) of radius.
start_pixel	Index of the first window's central pixel. If start_pixel is an int, it is used for all dimensions of data_array. To specify a custom starting element, pass a tuple with a step size for each dimension of data_array. If None, it defaults to radius, the moving window starts with a full slice.
step_size	Step size how far the window moves in one step. If step_size is an int, it is used for all dimensions of data_array. If None, it defaults to \(2r + 1\) for each element \(r\) of radius, which is equivalent to a rolling window.

Returns

Generator yielding in each iteration a tuple like (orig_pixel, target_pixel, window_content).

Return values

orig_pixel	Index of the window's center in data_array.
target_pixel	Index of the entry in a new array, in which the aggregated result is to be stored.
window_content	A view of the data_array, around the orig_pixel.

Definition at line 109 of file windows.py.

Here is the call graph for this function:

sliding()

fosanalysis.utils.windows.sliding	(	np.array	data_array,
			radius )

Generates a sliding window over an array.

This function returns a generator, hence, it should be use like:

for pixel, window in sliding(<array>, <radius>):
    # do something
    pass

In contrast to other function like numpy.lib.stride_tricks.sliding_window_view(), this window slides over each pixel, even in the margin areas of data_array. So, for each entry in data_array, a view to the window surrounding it is yielded. In the margins, the window contains fewer entries. Thus, boundary effects are to be expected, when using this function.

Remarks

Note, that a views of the original array are yielded, not copies. Changing a pixel's value in the window will change the original array.

Parameters

data_array	Array of data, over which the window should slide.
radius	Inradius of the window, sets the window's widths. Along an axis, the window has a width of \((2r+1)\). It is expected to be an int or iterable. If radius is an integer, it is used for all axes. The window will contain (up to) \((2r+1)^2\) entries. To set different widths for axes, use an iterable, such as a tuple. This tuple's length has to match the dimension of data_array.

Returns

In each iteration, a tuple like (pixel, window) is yielded.

Return values

pixel	Index of the window's central pixel in data_array.
window	Sub-array view of the data_array centered around pixel.

Definition at line 63 of file windows.py.

sliding_window_function()

np.array fosanalysis.utils.windows.sliding_window_function	(	np.array	arr,
			radius,
			fn,
		str	pad_mode = None,
		*	args,
		**	kwargs )

Applies the function fn to a sliding window over the array arr.

Note

Results in the margins (the first and last entries closer than radius to the edge of the array in each direction) are not reliable, as they suffer from boundary effects. This is caused by the sliding window only visit "complete" windows. In the margins, the values of the edge is repeated. The padding behavior can be changed with pad_mode.

Parameters

arr	Array of data, over which the window should slide.
radius	Inradius of the window, sets the window's widths. Along an axis, the window has a width of \((2r+1)\). It is expected to be an int a tuple. If radius is an integer, it is used for all axes. The windows will contain \((2r+1)^n\) entries. To set indiviual widths along for each dimension, use a tuple. This tuple's length has to match the dimension of arr.
fn	A function object (type: callable), taking a np.array as input and returning a float.
pad_mode	Mode for padding the edges of the result array. Defaults to "edge", which repeats the result value on the edge. For more, options, see numpy.pad()
*args	Additional positional arguments; ignored.
**kwargs	Additional keyword arguments; ignored.

Returns

Return a np.array with the same shape as arr. Each entry is the result of applying fn to a window reaching radius into each direction.

Definition at line 14 of file windows.py.