Finding linearly interdependent columns of a matrix in numpy - python

Problem: I have an MxN matrix where M>=N. I want to identify the groups of linearly-interdependent column-vectors within this matrix.
I'm hoping there's a fast and easy way to do this in numpy.
>>> a = np.random.randn(7, 6)
>>> a[:, 3] = 2*a[:, 0]-a[:, 4]
>>> a[:, 5] = 3*a[:, 1]
I'm looking for a function get_column_groups, which will return
>>> get_column_groups(a)
array([0, 1, 2, 0, 0, 1])
I guess bonus points if it also returns the rank of each group, e.g.:
>>> groups, group_ranks = get_column_groups(a)
>>> groups
array([0, 1, 2, 0, 0, 1])
>>> group_ranks
[2, 1, 1]

As far as I understand the question, computing the Spark of the matrix is a subproblem of what you are requesting, hence rendering it NP-Complete in most cases.
There may be some algorithms that do this better than just combinatorial evaluation of columns up to a size of rank(a), but this would be a start.

Related

Randomize numpy.argsort output in case of ties

I have a numpy array with some elements same as others i.e. there are ties, and I am applying np.argsort to find the indices which will sort the array:
In [29]: x = [1, 2, 1, 1, 5, 2]
In [30]: np.argsort(x)
Out[30]: array([0, 2, 3, 1, 5, 4])
In [31]: np.argsort(x)
Out[31]: array([0, 2, 3, 1, 5, 4])
As can be seen here, the outputs we get by running argsort two times are identical. However, array([2, 3, 0, 5, 1, 4]) is also a completely valid output because some elements in the original array are equal. Can I make argsort return me such "randomized" outputs when there are ties in my array? If not, what is a workaround because I don't want to bias my choice of the lowest values in the array when I am picking them.
One trick would be to add uniform noise in [0,1) range and then perform argsort-ing. Adding such a noise forces sorting only within their respective bins and gives randomized sort indices restricted to those bins -
(x+np.random.rand(len(x))).argsort()

Efficient way to create an array of nonzero values in a numpy tensor?

I have a 3D numpy grid A[ix,iy,iz] and I filter out elements by zeroing them out via A[ A<minval ] = 0, or A[ A>maxval ] = 0, etc. I then want to perform statistics on the remaining items. For now, I am doing:
for ai in np.reshape(A, nx*ny*nz):
if( ai > 0 ):
Atemp.append(ai)
and then I perform statistics on Atemp. This takes quite a long time, however, and I wonder if there is a more efficient way to create Atemp. For what it's worth, I am working with several GB of data in these arrays.
NOTE: I do not want a different way to filter these items out. I want to zero them out, then create a temporary array of all nonzero elements in A.
You can use:
Atemp = A[A != 0]
Eg:
In [3]: x = np.array([0,1,2,3,0,1,2,3,0]).reshape((3,3))
In [4]: x
Out[4]:
array([[0, 1, 2],
[3, 0, 1],
[2, 3, 0]])
In [5]: x[x == 0]
Out[5]: array([0, 0, 0])
In [6]: x[x != 0]
Out[6]: array([1, 2, 3, 1, 2, 3])
Another option:
Atemp = A.ravel()[np.flatnonzero(A)]

initialize a matrix in python of the row number

Is there a way to initialize a 3 row, 5 column matrix which contains these values without using a for loop?
[[0 0 0 0 0
1 1 1 1 1
2 2 2 2 2]]
It's possible.
i = 0
matrix = []
while i <=2:
matrix += [[i]*5]
i += 1
Without any for loops or list comprehensions, you can use a combination of built-in functions:
map(list, zip(*[range(3)] * 5))
If you're dealing with large datasets and are worried about performance, you might want to consider putting your data into a two-dimensional NumPy array. Here are a couple of ways of generating the matrix you ask for in NumPy:
>>> import numpy as np
>>> np.indices((3, 5))[0]
array([[0, 0, 0, 0, 0],
[1, 1, 1, 1, 1],
[2, 2, 2, 2, 2]])
>>> np.repeat(np.arange(3), 5).reshape((3, 5))
array([[0, 0, 0, 0, 0],
[1, 1, 1, 1, 1],
[2, 2, 2, 2, 2]])
The first of these is simpler, but a little bit wasteful: the np.indices call actually generates the array you want (which could be regarded as an array of row indices) along with a companion array of column indices:
>>> np.indices((3, 5))[1]
array([[0, 1, 2, 3, 4],
[0, 1, 2, 3, 4],
[0, 1, 2, 3, 4]])
with both arrays packed conveniently into a single array of shape (2, 3, 5). If you need that second array anyway for what you're doing then np.indices is the way to go (though in that case you may also want to look into NumPy's mgrid, ogrid and meshgrid functions). The second solution with np.repeat only generates the values you need, and not surprisingly, finishes about twice as fast on my machine when I bump the size of the matrix up to (3000, 5000):
In [19]: %timeit np.indices((3000, 5000))[0]
10 loops, best of 3: 156 ms per loop
In [20]: %timeit np.repeat(np.arange(3000), 5000).reshape((3000, 5000))
10 loops, best of 3: 88.4 ms per loop
Having said that, using np.repeat in this way is a little bit of an antipattern in NumPy: it's often better to avoid the repetition by creating a 2d array with 3 rows and a single column, and rely on NumPy's broadcasting to interpret this correctly when it's combined with other arrays. If you go that way, all you need is:
>>> np.arange(3)[:, np.newaxis]
array([[0],
[1],
[2]])
This is an array of shape (3, 1); a subsequent operation with an array of shape (5,) or (1, 5) (for example) would be subject to NumPy's broadcasting rules, producing an output of shape (3, 5). For example, here's what happens when we add a 1d array of zeros to the above:
>>> np.arange(3)[:, np.newaxis] + np.zeros(5, dtype=int)
array([[0, 0, 0, 0, 0],
[1, 1, 1, 1, 1],
[2, 2, 2, 2, 2]])
And for completeness, here's one more variation, using np.tile:
>>> np.tile(np.arange(3)[:, np.newaxis], (1, 5))
array([[0, 0, 0, 0, 0],
[1, 1, 1, 1, 1],
[2, 2, 2, 2, 2]])
All of these solutions should have reasonably similar performance for large values of 3 and 5; if this is a bottleneck, you'll want to do timings on your machine to decide which to use. On my machine, the +np.zeros broadcasting solution beats the others by some margin.
This is one of easy way to understand for Python Beginner.
matrix = []
for data in range(3):
matrix.append([data] * 5)
This is possible using:
[[data] * 5 for data in range(3)]

Scipy - find bases of column space of matrix

I'm trying to code up a simple Simplex algorithm, the first step of which is to find a basic feasible solution:
Choose a set B of linearly independent columns of A
Set all components of x corresponding to the columns not in B to zero.
Solve the m resulting equations to determine the components of x. These are the basic variables.
I know the solution will involve using scipy.linalg.svd (or scipy.linalg.lu) and some numpy.argwhere / numpy.where magic, but I'm not sure exactly how.
Does anyone have a pure-Numpy/Scipy implementation of finding a basis (step 1) or, even better, all of the above?
Example:
>>> A
array([[1, 1, 1, 1, 0, 0, 0],
[1, 0, 0, 0, 1, 0, 0],
[0, 0, 1, 0, 0, 1, 0],
[0, 3, 1, 0, 0, 0, 1]])
>>> u, s, v = scipy.linalg.svd(A)
>>> non_zero, = numpy.where(s > 1e-7)
>>> rank = len(non_zero)
>>> rank
4
>>> for basis in some_unknown_function(A):
... print(basis)
{3, 4, 5, 6}
{1, 4, 5, 6}
and so on.
A QR decomposition provides an orthogonal basis for the column space of A:
q,r = np.linalg.qr(A)
If the rank of A is n, then the first n columns of q form a basis for the column space of A.
Try using this
scipy.linalg.orth(A)
this produces orthonormal basis for the matrix A

How does the axis parameter from NumPy work?

Can someone explain exactly what the axis parameter in NumPy does?
I am terribly confused.
I'm trying to use the function myArray.sum(axis=num)
At first I thought if the array is itself 3 dimensions, axis=0 will return three elements, consisting of the sum of all nested items in that same position. If each dimension contained five dimensions, I expected axis=1 to return a result of five items, and so on.
However this is not the case, and the documentation does not do a good job helping me out (they use a 3x3x3 array so it's hard to tell what's happening)
Here's what I did:
>>> e
array([[[1, 0],
[0, 0]],
[[1, 1],
[1, 0]],
[[1, 0],
[0, 1]]])
>>> e.sum(axis = 0)
array([[3, 1],
[1, 1]])
>>> e.sum(axis=1)
array([[1, 0],
[2, 1],
[1, 1]])
>>> e.sum(axis=2)
array([[1, 0],
[2, 1],
[1, 1]])
>>>
Clearly the result is not intuitive.
Clearly,
e.shape == (3, 2, 2)
Sum over an axis is a reduction operation so the specified axis disappears. Hence,
e.sum(axis=0).shape == (2, 2)
e.sum(axis=1).shape == (3, 2)
e.sum(axis=2).shape == (3, 2)
Intuitively, we are "squashing" the array along the chosen axis, and summing the numbers that get squashed together.
To understand the axis intuitively, refer the picture below (source: Physics Dept, Cornell Uni)
The shape of the (boolean) array in the above figure is shape=(8, 3). ndarray.shape will return a tuple where the entries correspond to the length of the particular dimension. In our example, 8 corresponds to length of axis 0 whereas 3 corresponds to length of axis 1.
If someone need this visual description:
There are good answers for visualization however it might help to think purely from analytical perspective.
You can create array of arbitrary dimension with numpy.
For example, here's a 5-dimension array:
>>> a = np.random.rand(2, 3, 4, 5, 6)
>>> a.shape
(2, 3, 4, 5, 6)
You can access any element of this array by specifying indices. For example, here's the first element of this array:
>>> a[0, 0, 0, 0, 0]
0.0038908603263844155
Now if you take out one of the dimensions, you get number of elements in that dimension:
>>> a[0, 0, :, 0, 0]
array([0.00389086, 0.27394775, 0.26565889, 0.62125279])
When you apply a function like sum with axis parameter, that dimension gets eliminated and array of dimension less than original gets created. For each cell in new array, the operator will get list of elements and apply the reduction function to get a scaler.
>>> np.sum(a, axis=2).shape
(2, 3, 5, 6)
Now you can check that the first element of this array is sum of above elements:
>>> np.sum(a, axis=2)[0, 0, 0, 0]
1.1647502999560164
>>> a[0, 0, :, 0, 0].sum()
1.1647502999560164
The axis=None has special meaning to flatten out the array and apply function on all numbers.
Now you can think about more complex cases where axis is not just number but a tuple:
>>> np.sum(a, axis=(2,3)).shape
(2, 3, 6)
Note that we use same technique to figure out how this reduction was done:
>>> np.sum(a, axis=(2,3))[0,0,0]
7.889432081931909
>>> a[0, 0, :, :, 0].sum()
7.88943208193191
You can also use same reasoning for adding dimension in array instead of reducing dimension:
>>> x = np.random.rand(3, 4)
>>> y = np.random.rand(3, 4)
# New dimension is created on specified axis
>>> np.stack([x, y], axis=2).shape
(3, 4, 2)
>>> np.stack([x, y], axis=0).shape
(2, 3, 4)
# To retrieve item i in stack set i in that axis
Hope this gives you generic and full understanding of this important parameter.
Some answers are too specific or do not address the main source of confusion. This answer attempts to provide a more general but simple explanation of the concept, with a simple example.
The main source of confusion is related to expressions such as "Axis along which the means are computed", which is the documentation of the argument axis of the numpy.mean function. What the heck does "along which" even mean here? "Along which" essentially means that you will sum the rows (and divide by the number of rows, given that we are computing the mean), if the axis is 0, and the columns, if the axis is 1. In the case of axis is 0 (or 1), the rows can be scalars or vectors or even other multi-dimensional arrays.
In [1]: import numpy as np
In [2]: a=np.array([[1, 2], [3, 4]])
In [3]: a
Out[3]:
array([[1, 2],
[3, 4]])
In [4]: np.mean(a, axis=0)
Out[4]: array([2., 3.])
In [5]: np.mean(a, axis=1)
Out[5]: array([1.5, 3.5])
So, in the example above, np.mean(a, axis=0) returns array([2., 3.]) because (1 + 3)/2 = 2 and (2 + 4)/2 = 3. It returns an array of two numbers because it returns the mean of the rows for each column (and there are two columns).
Both 1st and 2nd reply is great for understanding ndarray concept in numpy. I am giving a simple example.
And according to this image by #debaonline4u
https://i.stack.imgur.com/O5hBF.jpg
Suppose , you have an 2D array -
[1, 2, 3]
[4, 5, 6]
In, numpy format it will be -
c = np.array([[1, 2, 3],
[4, 5, 6]])
Now,
c.ndim = 2 (rows/axis=0)
c.shape = (2,3) (axis0, axis1)
c.sum(axis=0) = [1+4, 2+5, 3+6] = [5, 7, 9] (sum of the 1st elements of each rows, so along axis0)
c.sum(axis=1) = [1+2+3, 4+5+6] = [6, 15] (sum of the elements in a row, so along axis1)
So for your 3D array,

Categories