We Keep Coding

Understanding the logic behind numpy code for Moore-Penrose inverse

I was going through the book called Hands-On Machine Learning with Scikit-Learn, Keras and Tensorflow and the author was explaining how the pseudo-inverse (Moore-Penrose inverse) of a matrix is calculated in the context of Linear Regression. I'm quoting verbatim here:
The pseudoinverse itself is computed using a standard matrix
factorization technique called Singular Value Decomposition (SVD) that
can decompose the training set matrix X into the matrix
multiplication of three matrices U Σ VT (see numpy.linalg.svd()). The
pseudoinverse is calculated as X+ = V * Σ+ * UT. To compute the matrix
Σ+, the algorithm takes Σ and sets to zero all values smaller than a
tiny threshold value, then it replaces all nonzero values with their
inverse, and finally it transposes the resulting matrix. This approach
is more efficient than computing the Normal equation.
I've got an understanding of how the pseudo-inverse and SVD are related from this post. But I'm not able to grasp the rationale behind setting all values less than the threshold to zero. The inverse of a diagonal matrix is obtained by taking the reciprocals of the diagonal elements. Then small values would be converted to large values in the inverse matrix, right? Then why are we removing the large values?
I went and looked into the numpy code, and it looks like follows, just for reference:
#array_function_dispatch(_pinv_dispatcher)
def pinv(a, rcond=1e-15, hermitian=False):
a, wrap = _makearray(a)
rcond = asarray(rcond)
if _is_empty_2d(a):
m, n = a.shape[-2:]
res = empty(a.shape[:-2] + (n, m), dtype=a.dtype)
return wrap(res)
a = a.conjugate()
u, s, vt = svd(a, full_matrices=False, hermitian=hermitian)
# discard small singular values
cutoff = rcond[..., newaxis] * amax(s, axis=-1, keepdims=True)
large = s > cutoff
s = divide(1, s, where=large, out=s)
s[~large] = 0
res = matmul(transpose(vt), multiply(s[..., newaxis], transpose(u)))
return wrap(res)

It's almost certainly an adjustment for numerical error. To see why this might be necessary, look what happens when you take the svd of a rank-one 2x2 matrix. We can create a rank-one matrix by taking the outer product of a vector like so:
>>> a = numpy.arange(2) + 1
>>> A = a[:, None] * a[None, :]
>>> A
array([[1, 2],
[2, 4]])
Although this is a 2x2 matrix, it only has one linearly independent column, and so its rank is one instead of two. So we should expect that when we pass it to svd, one of the singular values will be zero. But look what happens:
>>> U, s, V = numpy.linalg.svd(A)
>>> s
array([5.00000000e+00, 1.98602732e-16])
What we actually get is a singular value that is not quite zero. This result is inevitable in many cases given that we are working with finite-precision floating point numbers. So although the problem you have identified is a real one, we will not be able to tell in practice the difference between a matrix that really has a very small singular value and a matrix that ought to have a zero singular value but doesn't. Setting small values to zero is the safest practical way to handle that problem.

Why multiplication functions of scipy sparse and numpy arrays give different results?

I have two matrices in Python 2.7: one dense A_dense and the another sparse matrix A_sparse. I am interested in computing element-wise multiplication followed by sum. There are two ways to do it: use numpy's multiplication or scipy sparse multiplication. I expect them to give exactly same result with difference in execution time. But I find that they give different results for certain matrix sizes.
import numpy as np
from scipy import sparse
L=2000
np.random.seed(2)
rand_x=np.random.rand(L)
A_sparse_init=np.diag(rand_x, -1)+np.diag(rand_x, 1)
A_sparse=sparse.csr_matrix(A_sparse_init)
A_dense=np.random.rand(L+1,L+1)
print np.sum(A_sparse.multiply(A_dense))-np.sum(np.multiply(A_dense[A_sparse.nonzero()], A_sparse.data))
Output:
1.1368683772161603e-13
If I choose L=2001, then output is:
0.0
To check the size dependence of the difference using two different multiplication method, I wrote:
L=100
np.random.seed(2)
N_loop=100
multiply_diff_arr=np.zeros(N_loop)
for i in xrange(N_loop):
rand_x=np.random.rand(L)
A_sparse_init=np.diag(rand_x, -1)+np.diag(rand_x, 1)
A_sparse=sparse.csr_matrix(A_sparse_init)
A_dense=np.random.rand(L+1,L+1)
multiply_diff_arr[i]=np.sum(A_sparse.multiply(A_dense))-np.sum(np.multiply(A_dense[A_sparse.nonzero()], A_sparse.data))
L+=1
I got the following plot:
Can anyone help me understand what's happening? Don't we expect the difference between two methods to be at least 1e-18 rather than 1e-13?

I don't have a full answer, but this might help find the answer:
Under the hood, scipy.sparse will convert to coo format and do this:
ret = self.tocoo()
if self.shape == other.shape:
data = np.multiply(ret.data, other[ret.row, ret.col])
The question is then why these two operations give different results:
ret = A_sparse.tocoo()
c = np.multiply(ret.data, A_dense[ret.row, ret.col])
ret.data = c.view(type=np.ndarray)
c.sum() - ret.sum()
-1.1368683772161603e-13
Edit:
The difference stems from different defaults on which axis to add.reduce first.
E.g.:
A_sparse.multiply(A_dense).sum(axis=1).sum()
A_sparse.multiply(A_dense).sum(axis=0).sum()
Numpy defaults to 0 first.

Can covariance of A be used to calculate A'*A?

I am doing a benchmarking test in python on different ways to calculate A'*A, with A being a N x M matrix. One of the fastest ways was to use numpy.dot().
I was curious if I can obtain the same result using numpy.cov() (which gives the covariance matrix) by somehow varying the weights or by somehow pre-processing the A matrix ? But I had no success. Does anyone know if there is any relation between the product A'*A and covariance of A, where A is a matrix with N rows/observations and M columns/variables?

Have a look at the cov source. Near the end of the function it does this:
c = dot(X, X_T.conj())
Which is basically the dot product you want to perform. However, there are all kinds of other operations: checking inputs, subtracting the mean, normalization, ...
In short, np.cov will never ever be faster than np.dot(A.T, A) because internally it contains exactly that operation.
That said - the covariance matrix is computed as
Or in Python:
import numpy as np
a = np.random.rand(10, 3)
m = np.mean(a, axis=0, keepdims=True)
x = np.dot((a - m).T, a - m) / (a.shape[0] - 1)
y = np.cov(a.T)
assert np.allclose(x, y) # check they are equivalent
As you can see, the covariance matrix is equivalent to the raw dot product if you subtract the mean of each variable and divide the result by the number of samples (minus one).

Matrix inverse in numpy/python not giving correct matrix?

I have a nxn matrix C and use inv from numpy.linalg to take the inverse to get Cinverse. My Cmatrix has elements of order 10**4 but my Cinverse matrix has elements of order 10**12 and higher (not sure if thats correct). When I do numpyp.dot(C,Cinverse), I do not get the identity matrix. Why is this?
I have a vector x which I multiply by itself to get a matrix.
x=array([ 121.41191662, 74.22830468, 73.23156336, 75.48354975,
79.89580817])
c=np.outer(xvector,xvector)
this is a 5x5 matrix.
then I get its inverse by
from numpy.linalg import inv
cinverse=inv(c)
then I want to see if I can get identity matrix back.
identity=np.dot(C00,C00inv)
However, I do not get the identity matrix. cinverse has very large matrix elements
around 10**13 and higher while c has matrix elements around 10,000.

The outer product of two vectors (be they the same or not) is not invertible. Since it is just a stack of scaled copies of the same vector its rank is one. Rank defective matrices cannot be inverted.
I'm surprised that numpy is not raising an exception or at least giving a warning.

So here is some code that generates the inverse matrix, and I will comment about it afterwards.
import numpy as np
x = np.random.rand(5,5)*10000 # makes a 5x5 matrix with elements around 10000
xin = np.linalg.inv(x)
iden = np.dot(x,xinv)
Now the first line of your iden matrix probably looks something like this:
[ 1.00000000e+00, -2.05382445e-16, -5.61067365e-16, 1.99719718e-15, -2.12322957e-16]
. Notice that the first element is exactly 1, as it should be, but there others are not exactly 0, however they are essentially zero and should be regarded as zero according to machine precision.

What's wrong with my PCA?

My code:
from numpy import *
def pca(orig_data):
data = array(orig_data)
data = (data - data.mean(axis=0)) / data.std(axis=0)
u, s, v = linalg.svd(data)
print s #should be s**2 instead!
print v
def load_iris(path):
lines = []
with open(path) as input_file:
lines = input_file.readlines()
data = []
for line in lines:
cur_line = line.rstrip().split(',')
cur_line = cur_line[:-1]
cur_line = [float(elem) for elem in cur_line]
data.append(array(cur_line))
return array(data)
if __name__ == '__main__':
data = load_iris('iris.data')
pca(data)
The iris dataset: http://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data
Output:
[ 20.89551896 11.75513248 4.7013819 1.75816839]
[[ 0.52237162 -0.26335492 0.58125401 0.56561105]
[-0.37231836 -0.92555649 -0.02109478 -0.06541577]
[ 0.72101681 -0.24203288 -0.14089226 -0.6338014 ]
[ 0.26199559 -0.12413481 -0.80115427 0.52354627]]
Desired Output:
Eigenvalues - [2.9108 0.9212 0.1474 0.0206]
Principal Components - Same as I got but transposed so okay I guess
Also, what's with the output of the linalg.eig function? According to the PCA description on wikipedia, I'm supposed to this:
cov_mat = cov(orig_data)
val, vec = linalg.eig(cov_mat)
print val
But it doesn't really match the output in the tutorials I found online. Plus, if I have 4 dimensions, I thought I should have 4 eigenvalues and not 150 like the eig gives me. Am I doing something wrong?
edit: I've noticed that the values differ by 150, which is the number of elements in the dataset. Also, the eigenvalues are supposed to add to be equal to the number of dimensions, in this case, 4. What I don't understand is why this difference is happening. If I simply divided the eigenvalues by len(data) I could get the result I want, but I don't understand why. Either way the proportion of the eigenvalues isn't altered, but they are important to me so I'd like to understand what's going on.

You decomposed the wrong matrix.
Principal Component Analysis requires manipulating the eigenvectors/eigenvalues
of the covariance matrix, not the data itself. The covariance matrix, created from an m x n data matrix, will be an m x m matrix with ones along the main diagonal.
You can indeed use the cov function, but you need further manipulation of your data. It's probably a little easier to use a similar function, corrcoef:
import numpy as NP
import numpy.linalg as LA
# a simulated data set with 8 data points, each point having five features
data = NP.random.randint(0, 10, 40).reshape(8, 5)
# usually a good idea to mean center your data first:
data -= NP.mean(data, axis=0)
# calculate the covariance matrix
C = NP.corrcoef(data, rowvar=0)
# returns an m x m matrix, or here a 5 x 5 matrix)
# now get the eigenvalues/eigenvectors of C:
eval, evec = LA.eig(C)
To get the eigenvectors/eigenvalues, I did not decompose the covariance matrix using SVD,
though, you certainly can. My preference is to calculate them using eig in NumPy's (or SciPy's)
LA module--it is a little easier to work with than svd, the return values are the eigenvectors
and eigenvalues themselves, and nothing else. By contrast, as you know, svd doesn't return these these directly.
Granted the SVD function will decompose any matrix, not just square ones (to which the eig function is limited); however when doing PCA, you'll always have a square matrix to decompose,
regardless of the form that your data is in. This is obvious because the matrix you
are decomposing in PCA is a covariance matrix, which by definition is always square
(i.e., the columns are the individual data points of the original matrix, likewise
for the rows, and each cell is the covariance of those two points, as evidenced
by the ones down the main diagonal--a given data point has perfect covariance with itself).

The left singular values returned by SVD(A) are the eigenvectors of AA^T.
The covariance matrix of a dataset A is : 1/(N-1) * AA^T
Now, when you do PCA by using the SVD, you have to divide each entry in your A matrix by (N-1) so you get the eigenvalues of the covariance with the correct scale.
In your case, N=150 and you haven't done this division, hence the discrepancy.
This is explained in detail here

(Can you ask one question, please? Or at least list your questions separately. Your post reads like a stream of consciousness because you are not asking one single question.)
You probably used cov incorrectly by not transposing the matrix first. If cov_mat is 4-by-4, then eig will produce four eigenvalues and four eigenvectors.
Note how SVD and PCA, while related, are not exactly the same. Let X be a 4-by-150 matrix of observations where each 4-element column is a single observation. Then, the following are equivalent:
a. the left singular vectors of X,
b. the principal components of X,
c. the eigenvectors of X X^T.
Also, the eigenvalues of X X^T are equal to the square of the singular values of X. To see all this, let X have the SVD X = QSV^T, where S is a diagonal matrix of singular values. Then consider the eigendecomposition D = Q^T X X^T Q, where D is a diagonal matrix of eigenvalues. Replace X with its SVD, and see what happens.

Question already adressed: Principal component analysis in Python