假设我有以下数组:
a = np.array([[1,2,3,4,5,6],[7,8,9,10,11,12],[3,6,7,9]])
我想对第一行的前两个值求和:1 2 = 3,然后是下两个值:3 4 = 7,然后5 6 = 11,依此类推每一行.我想要的输出是这样的:
array([[ 3,11],[15,19,23],[ 8,13,17]])
我有以下解决方案:
def sum_chunks(x,chunk_size): rows,cols = x.shape x = x.reshape(x.size / chunk_size,chunk_size) return x.sum(axis=1).reshape(rows,cols/chunk_size)
但感觉不必要的复杂,有更好的方法吗?也许内置?
解决方法
当我必须做这种事情时,我更喜欢将2D阵列重塑为3D阵列,然后使用np.sum折叠额外的维度.将其推广到n维数组,你可以这样做:
def sum_chunk(x,chunk_size,axis=-1): shape = x.shape if axis < 0: axis += x.ndim shape = shape[:axis] + (-1,chunk_size) + shape[axis+1:] x = x.reshape(shape) return x.sum(axis=axis+1) >>> a = np.arange(24).reshape(4,6) >>> a array([[ 0,1,5],[ 6,[12,14,15,16,17],[18,20,21,22,23]]) >>> sum_chunk(a,2) array([[ 1,9],[13,17,21],[25,29,33],[37,41,45]]) >>> sum_chunk(a,axis=0) array([[ 6,12,16],[30,32,34,36,38,40]])