我试图为我的排序numpy数组创建一个diff,这样如果我记录第一行的值和diff,就可以重新创建原始表,但是存储的数据更少。在
下面是一个表格的例子:my_array = numpy.array([(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0),
(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1),
(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2),
(9, 36, 146, 73, 36, 146, 73, 36, 146, 73, 36, 146, 73, 34),
(9, 36, 146, 73, 36, 146, 73, 36, 146, 73, 36, 146, 73, 35),
(9, 36, 146, 73, 36, 146, 73, 36, 146, 73, 36, 146, 73, 36)
],'uint8,uint8,uint8,uint8,uint8,uint8,uint8,uint8,uint8,uint8,uint8,uint8,uint8,uint8')
跑完之后数字差异(my峎u array)我会预料到这样的情况:
^{pr2}$
Note: The data above comes from the first & last three rows of the
'real' data, which is much much larger. With the full dataset, most of the
rows after a diff would be 0,0,0,0,0,0,0,0,0,0,0,0,1 -- which can a)
be stored in a much smaller struct, and b) will compress fantastically well on disk since most rows contain very similar data.
I should probably point out that the reason I have a whole bunch of uint8's in the first place, is because I needed to store an array of extremely large numbers, in the smallest amount of memory possible. The largest number was 185439173519100986733232011757860, which is too big for uint64. In fact, the smallest number of bits to store it would be 108 bits, or 14 bytes (to the nearest byte). So to fit these large numbers into numpy, i use the following two functions:
def large_number_to_numpy(number,columns):
return tuple((number >> (8*x)) & 255 for x in range(columns-1,-1,-1))
def numpy_to_large_number(numbers):
return sum([y << (8*x) for x,y in enumerate(numbers[::-1])])
Which is used like this:
>>> large_number_to_numpy(185439173519100986733232011757860L,14)
(9L, 36L, 146L, 73L, 36L, 146L, 73L, 36L, 146L, 73L, 36L, 146L, 73L, 36L)
numpy_to_large_number((9L, 36L, 146L, 73L, 36L, 146L, 73L, 36L, 146L, 73L, 36L, 146L, 73L, 36L))
185439173519100986733232011757860L
With the array created like this:
my_array = numpy.zeros(TOTAL_ROWS,','.join(14*['uint8']))
And then populated with:
my_array[x] = large_number_to_numpy(large_number,14)
但我得到的却是:>>> my_array
array([(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0),
(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1),
(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2),
(9, 36, 146, 73, 36, 146, 73, 36, 146, 73, 36, 146, 73, 34),
(9, 36, 146, 73, 36, 146, 73, 36, 146, 73, 36, 146, 73, 35),
(9, 36, 146, 73, 36, 146, 73, 36, 146, 73, 36, 146, 73, 36)],
dtype=[('f0', 'u1'), ('f1', 'u1'), ('f2', 'u1'), ('f3', 'u1'), ('f4', 'u1'), ('f5', 'u1'), ('f6', 'u1'), ('f7', 'u1'), ('f8', 'u1'), ('f9', 'u1'), ('f10', 'u1'), ('f11', 'u1'), ('f12', 'u1'), ('f13', 'u1')])
>>> numpy.diff(my_array)
Traceback (most recent call last):
File "", line 1, in
File "/usr/local/lib/python2.7/site-packages/numpy/lib/function_base.py", line 1567, in diff
return a[slice1]-a[slice2]
TypeError: ufunc 'subtract' did not contain a loop with signature matching types dtype([('f0', 'u1'), ('f1', 'u1'), ('f2', 'u1'), ('f3', 'u1'), ('f4', 'u1'), ('f5', 'u1'), ('f6', 'u1'), ('f7', 'u1'), ('f8', 'u1'), ('f9', 'u1'), ('f10', 'u1'), ('f11', 'u1'), ('f12', 'u1'), ('f13', 'u1')]) dtype([('f0', 'u1'), ('f1', 'u1'), ('f2', 'u1'), ('f3', 'u1'), ('f4', 'u1'), ('f5', 'u1'), ('f6', 'u1'), ('f7', 'u1'), ('f8', 'u1'), ('f9', 'u1'), ('f10', 'u1'), ('f11', 'u1'), ('f12', 'u1'), ('f13', 'u1')]) dtype([('f0', 'u1'), ('f1', 'u1'), ('f2', 'u1'), ('f3', 'u1'), ('f4', 'u1'), ('f5', 'u1'), ('f6', 'u1'), ('f7', 'u1'), ('f8', 'u1'), ('f9', 'u1'), ('f10', 'u1'), ('f11', 'u1'), ('f12', 'u1'), ('f13', 'u1')])