Question

如果我有一个数组x，一个起始索引数组s和一个距离d，我该如何创建一个数组y，s.t。 y[i] == x[s[i]:s[i]+d]？

Answer 1

您可以创建滑动d - 窗口的纵向视图到x，然后使用高级索引：

>>> x = np.random.randint(0, 10, (20,))
>>> d = 4
>>> s = np.random.randint(0, 17, (10,))
>>> 
>>> x
array([2, 4, 5, 6, 6, 4, 8, 2, 6, 6, 0, 2, 8, 3, 5, 1, 6, 2, 9, 8])
>>> s
array([16,  4,  6,  0, 16,  1, 14,  4,  6,  6])
>>> 
>>> assert x.ndim == 1
>>> assert x.flags.contiguous
>>>
>>> xx  = np.lib.stride_tricks.as_strided(x, (x.size-d+1, d), 2*x.strides)
>>> y = xx[s]
>>> y
array([[6, 2, 9, 8],
       [6, 4, 8, 2],
       [8, 2, 6, 6],
       [2, 4, 5, 6],
       [6, 2, 9, 8],
       [4, 5, 6, 6],
       [5, 1, 6, 2],
       [6, 4, 8, 2],
       [8, 2, 6, 6],
       [8, 2, 6, 6]])

Answer 2

不确定这是否是最佳方式，但您可以使用元组索引：

select CAST(timestamp AS DATE) as date, execution_id, min(timestamp) as begin_timestamp, max(timestamp) as end_timestamp, new_block_id, 
sum(case when Type = 'P' then 1 else 0 end) as P_count,
              sum(case when Type = 'D' then 1 else 0 end) as D_count
from (
    select *,
    max(block_id) over (partition by EmpID order by timestamp ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) new_block_id  from myTable
) d
group by EmpID, new_block_id, CAST(timestamp AS DATE)
order by EmpID, new_block_id, CAST(timestamp AS DATE)

numpy在开始和结束索引之间获取数组

2 个答案: