How to tile a 1D numpy array using uneven subarrays as tiles?

Question

Is there a way to use sub-arrays of a 1-D array as the input tiles for np.tile? I start with:

• a 1D array,
• the sizes of each of the tiles,
• the number of repeats for each tile.

In this case, the number of repeats for each tile is equal to the number of elements in that tile.

Example:

arr = np.array([0,1,2,3,4])
tile_sizes = np.array([2, 3])
num_repeats = tile_sizes

#do some np.tile thing here

and the output array will be:

np.array([0,1,0,1,2,3,4,2,3,4,2,3,4])

note that the first 2 elements (0 and 1) formed a tile of shape (2,) which was repeated 2 times. The next tile was 3 elements (2,3, and 4) and was tiled 3 times.

The use-case for this will involve arrays of a million elements, so memory and speed are concerns, meaning broadcasting is preferred.

A non-broadcasting way to achieve this looks like:

tiles = np.split(arr, np.cumsum(tile_sizes)[:-1])
repeated_tiles = [np.tile(tile, tile.shape[0]) for tile in tiles]
output = np.concatenate(repeated_tiles)
output
>>>>>
array([0, 1, 0, 1, 2, 3, 4, 2, 3, 4, 2, 3, 4])

Answer 1

It's not a perfect solution, but you can get rid of the list comprehension using np.repeat if that helps.

a = np.arange(5)
tile_sizes = np.array([2, 3])
tiles = np.array(np.split(a, np.cumsum(tile_sizes)[:-1]), dtype=np.object)
tiles = np.concatenate(np.repeat(tiles, tile_sizes))