SIice a numpy array based on an interval? - numpy

Is it possible to systematically slice an 1d array of length m by an interval n in numpy? Say I have a list of 1000 values, could I break that into 10 lists of 100 values easily?

You can use both np.array_split() and np.split() which in fact are the same with a little note (as per np.array_split())
From the documentation:
x = np.arange(8.0)
np.array_split(x, 3)
#Result
[array([0., 1., 2.]), array([3., 4., 5.]), array([6., 7.])]
Split an array into multiple sub-arrays.
Please refer to the split documentation. The only difference between
these functions is that array_split allows indices_or_sections to be
an integer that does not equally divide the axis. For an array of
length l that should be split into n sections, it returns l % n
sub-arrays of size l//n + 1 and the rest of size l//n.

array_split allows one to split with unequal spacing as well, should this ever meet your needs
ar = np.arange(0, 20, dtype='int')
s = [2, 7, 12, 17]
np.array_split(ar, s)
Out[80]:
[array([0, 1]),
array([2, 3, 4, 5, 6]),
array([ 7, 8, 9, 10, 11]),
array([12, 13, 14, 15, 16]),
array([17, 18, 19])]

Related

Numpy subarrays and relative indexing

I have been searching if there is an standard mehtod to create a subarray using relative indexes. Take the following array into consideration:
>>> m = np.arange(25).reshape([5, 5])
array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14],
[15, 16, 17, 18, 19],
[20, 21, 22, 23, 24]])
I want to access the 3x3 matrix at a specific array position, for example [2,2]:
>>> x = 2, y = 2
>>> m[slice(x-1,x+2), slice(y-1,y+2)]
array([[ 6, 7, 8],
[11, 12, 13],
[16, 17, 18]])
For example for the above somethig like m.subarray(pos=[2,2], shape=[3,3])
I want to sample a ndarray of n dimensions on a specific position which might change.
I did not want to use a loop as it might be inneficient. Scipy functions correlate and convolve do this very efficiently, but for all positions. I am interested only in the sampling of one.
The best answer could solve the issues at edges, in my case I would like for example to have wrap mode:
(a b c d | a b c d | a b c d)
--------------------EDITED-----------------------------
Based on the answer from #Carlos Horn, I could create the following function.
def cell_neighbours(array, index, shape):
pads = [(floor(dim/2), ceil(dim / 2)) for dim in shape]
array = np.pad(self.configuration, pads, "wrap")
views = np.lib.stride_tricks.sliding_window_view
return views(array, shape)[tuple(index)]
Last concern might be about speed, from docs: For many applications using a sliding window view can be convenient, but potentially very slow. Often specialized solutions exist.
From here maybe is easier to get a faster solution.
You could build a view of 3x3 matrices into the array as follows:
import numpy as np
m = np.arange(25).reshape(5,5)
m3x3view = np.lib.stride_tricks.sliding_window_view(m, (3,3))
Note that it will change slightly your indexing on half the window size meaning
x_view = x - 3//2
y_view = y - 3//2
print(m3x3view[x_view,y_view]) # gives your result
In case a copy operation is fine, you could use:
mpad = np.pad(m, 1, mode="wrap")
mpad3x3view = np.lib.stride_tricks.sliding_window_view(mpad, (3,3))
print(mpad3x3view[x % 5,y % 5])
to use arbitrary x, y integer values.

Numpy Interpolation for Array of Arrays

I have an array of arrays that I want to interpolate based on each array's min and max.
For a simple mxn array , with values ranging from 0 to 1, I can do this as follows:
x_inp=np.interp(x,(x.min(),x.max()),(0,0.7))
This suppresses every existing value to 0 to 0.7. However, if I have an array of dimension 100xmxn, the above method considers the global min/max and not the individual min/max of each of the mxn array.
Edit:
For example
x1=np.random.randint(0,5, size=(2, 4))
x2=np.random.randint(6,10, size=(2, 4))
my_list=[x1,x2]
my_array=np.asarray(my_list)
print(my_array)
>> array([[[1, 4, 3, 4],
[3, 2, 0, 0]],
[9, 6, 8, 6],
8, 7, 6, 7]]])
my_array is now of dimension 2x2x4 and my_array.min() and my_array.max() would give me 0 and 9. So If I interpolate, it won't work based on the min/max of the individual 2x4 arrays. What I want is, to have the interpolation work based on min/max of 0/4 for the 1st array and 6/9 for the second.

numpy find values of maxima pointed to by argmax [duplicate]

This question already has answers here:
Index n dimensional array with (n-1) d array
(3 answers)
Closed 4 years ago.
I have a 3-d array. I find the indexes of the maxima along an axis using argmax. How do I now use these indexes to obtain the maximal values?
2nd part: How to do this for arrays of N-d?
Eg:
u = np.arange(12).reshape(3,4,1)
In [125]: e = u.argmax(axis=2)
Out[130]: e
array([[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]])
It would be nice if u[e] produced the expected results, but it doesn't work.
The return value of argmax along an axis can't be simply used as an index. It only works in a 1d case.
In [124]: u = np.arange(12).reshape(3,4,1)
In [125]: e = u.argmax(axis=2)
In [126]: u.shape
Out[126]: (3, 4, 1)
In [127]: e.shape
Out[127]: (3, 4)
e is (3,4), but its values only index the last dimension of u.
In [128]: u[e].shape
Out[128]: (3, 4, 4, 1)
Instead we have to construct indices for the other 2 dimensions, ones which broadcast with e. For example:
In [129]: I,J=np.ix_(range(3),range(4))
In [130]: I
Out[130]:
array([[0],
[1],
[2]])
In [131]: J
Out[131]: array([[0, 1, 2, 3]])
Those are (3,1) and (1,4). Those are compatible with (3,4) e and the desired output
In [132]: u[I,J,e]
Out[132]:
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]])
This kind of question has been asked before, so probably should be marked as a duplicate. The fact that your last dimension is size 1, and hence e is all 0s, distracting readers from the underlying issue (using a multidimensional argmax as index).
numpy: how to get a max from an argmax result
Get indices of numpy.argmax elements over an axis
Assuming you've taken the argmax on the last dimension
In [156]: ij = np.indices(u.shape[:-1])
In [157]: u[(*ij,e)]
Out[157]:
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]])
or:
ij = np.ix_(*[range(i) for i in u.shape[:-1]])
If the axis is in the middle, it'll take a bit more tuple fiddling to arrange the ij elements and e.
so for general N-d array
dims = np.ix_(*[range(x) for x in u.shape[:-1]])
u.__getitem__((*dims,e))
You can't write u[*dims,e], that's a syntax error, so I think you must use getitem directly.

Find indexes of nonuniform sample with np.random.choice

Let's say I have a positions information in the form a two large 1D arrays X and Y. I want to sample non-uniformly positions from these arrays.
I thought I could do this with np.random.choice, but since it only accepts 1D arrays and I cannot do:
Xsample = np.random.choice(X, n, p)
Ysample = np.random.choice(Y, n, p)
with n number of points in the sample, and p a probability array, since this will sample different points for Xsample and Ysample, I am left with finding a way to obtain the indexes of one sampling. The problem is that there is no guarantee that the numbers in the lists are unique so cannot quite use np.where.
Any thoughts?
Doh, I can just sample from the indexes.
Here's a working example:
X = np.array([1, 2, 3, 4, 5])
Y = np.array([11, 12, 13, 14, 15])
p = [0.25, 0., 0.5, 0.25]
sample_idxs = np.random.choice(arange(len(X)), 2, p)
# can also be
# sample_idxs = np.random.choice(len(X), 2, p)
sample_idxs
> array([2, 4])
X[sample_idxs]
> array([3, 5])
Y[sample_idxs]
> array([13, 15])

Numpy rebinning a 2D array

I am looking for a fast formulation to do a numerical binning of a 2D numpy array. By binning I mean calculate submatrix averages or cumulative values. For ex. x = numpy.arange(16).reshape(4, 4) would have been splitted in 4 submatrix of 2x2 each and gives numpy.array([[2.5,4.5],[10.5,12.5]]) where 2.5=numpy.average([0,1,4,5]) etc...
How to perform such an operation in an efficient way... I don't have really any ideay how to perform this ...
Many thanks...
You can use a higher dimensional view of your array and take the average along the extra dimensions:
In [12]: a = np.arange(36).reshape(6, 6)
In [13]: a
Out[13]:
array([[ 0, 1, 2, 3, 4, 5],
[ 6, 7, 8, 9, 10, 11],
[12, 13, 14, 15, 16, 17],
[18, 19, 20, 21, 22, 23],
[24, 25, 26, 27, 28, 29],
[30, 31, 32, 33, 34, 35]])
In [14]: a_view = a.reshape(3, 2, 3, 2)
In [15]: a_view.mean(axis=3).mean(axis=1)
Out[15]:
array([[ 3.5, 5.5, 7.5],
[ 15.5, 17.5, 19.5],
[ 27.5, 29.5, 31.5]])
In general, if you want bins of shape (a, b) for an array of (rows, cols), your reshaping of it should be .reshape(rows // a, a, cols // b, b). Note also that the order of the .mean is important, e.g. a_view.mean(axis=1).mean(axis=3) will raise an error, because a_view.mean(axis=1) only has three dimensions, although a_view.mean(axis=1).mean(axis=2) will work fine, but it makes it harder to understand what is going on.
As is, the above code only works if you can fit an integer number of bins inside your array, i.e. if a divides rows and b divides cols. There are ways to deal with other cases, but you will have to define the behavior you want then.
See the SciPy Cookbook on rebinning, which provides this snippet:
def rebin(a, *args):
'''rebin ndarray data into a smaller ndarray of the same rank whose dimensions
are factors of the original dimensions. eg. An array with 6 columns and 4 rows
can be reduced to have 6,3,2 or 1 columns and 4,2 or 1 rows.
example usages:
>>> a=rand(6,4); b=rebin(a,3,2)
>>> a=rand(6); b=rebin(a,2)
'''
shape = a.shape
lenShape = len(shape)
factor = asarray(shape)/asarray(args)
evList = ['a.reshape('] + \
['args[%d],factor[%d],'%(i,i) for i in range(lenShape)] + \
[')'] + ['.sum(%d)'%(i+1) for i in range(lenShape)] + \
['/factor[%d]'%i for i in range(lenShape)]
print ''.join(evList)
return eval(''.join(evList))
I assume that you only want to know how to generally build a function that performs well and does something with arrays, just like numpy.reshape in your example. So if performance really matters and you're already using numpy, you can write your own C code for that, like numpy does. For example, the implementation of arange is completely in C. Almost everything with numpy which matters in terms of performance is implemented in C.
However, before doing so you should try to implement the code in python and see if the performance is good enough. Try do make the python code as efficient as possible. If it still doesn't suit your performance needs, go the C way.
You may read about that in the docs.