I've come across a from scratch implementation for gaussian processes:
http://krasserm.github.io/2018/03/19/gaussian-processes/
There, the isotropic squared exponential kernel is implemented in numpy. It looks like:
The implementation is:
def kernel(X1, X2, l=1.0, sigma_f=1.0):
sqdist = np.sum(X1**2, 1).reshape(-1, 1) + np.sum(X2**2, 1) - 2 * np.dot(X1, X2.T)
return sigma_f**2 * np.exp(-0.5 / l**2 * sqdist)
consistent with the implementation of Nando de Freitas: https://www.cs.ubc.ca/~nando/540-2013/lectures/gp.py
However, I am not quite sure how this implementation matches the provided formula, especially in the sqdist part. In my opinion, it is wrong but it works (and delivers the same results as scipy's cdist with squared euclidean distance). Why do I think it is wrong? If you multiply out the multiplication of the two matrices, you get
which equals to either a scalar or a nxn matrix for a vector x_i, depending on whether you define x_i to be a column vector or not. The implementation however gives back a nx1 vector with the squared values.
I hope that anyone can shed light on this.
I found out: The implementation is correct. I just was not aware of the fuzzy notation (in my opinion) which is sometimes used in ML contexts. What is to be achieved is a distance matrix and each row vectors of matrix A are to be compared with the row vectors of matrix B to infer the covariance matrix, not (as I somehow guessed) the direct distance between two matrices/vectors.
Related
I'm new to automatic differentiation programming, so this maybe a naive question. Below is a simplified version of what I'm trying to solve.
I have two input arrays - a vector A of size N and a matrix B of shape (N, M), as well a parameter vector theta of size M. I define a new array C(theta) = B * theta to get a new vector of size N. I then obtain the indices of elements that fall in the upper and lower quartile of C, and use them to create a new array A_low(theta) = A[lower quartile indices of C] and A_high(theta) = A[upper quartile indices of C]. Clearly these two do depend on theta, but is it possible to differentiate A_low and A_high w.r.t theta?
My attempts so far seem to suggest no - I have using the python libraries of autograd, JAX and tensorflow, but they all return a gradient of zero. (The approaches I have tried so far involve using argsort or extracting the relevant sub-arrays using tf.top_k.)
What I'm seeking help with is either a proof that the derivative is not defined (or cannot be analytically computed) or if it does exist, a suggestion on how to estimate it. My eventual goal is to minimize some function f(A_low, A_high) wrt theta.
This is the JAX computation that I wrote based on your description:
import numpy as np
import jax.numpy as jnp
import jax
N = 10
M = 20
rng = np.random.default_rng(0)
A = jnp.array(rng.random((N,)))
B = jnp.array(rng.random((N, M)))
theta = jnp.array(rng.random(M))
def f(A, B, theta, k=3):
C = B # theta
_, i_upper = lax.top_k(C, k)
_, i_lower = lax.top_k(-C, k)
return A[i_lower], A[i_upper]
x, y = f(A, B, theta)
dx_dtheta, dy_dtheta = jax.jacobian(f, argnums=2)(A, B, theta)
The derivatives are all zero, and I believe this is correct, because the change in value of the outputs does not depend on the change in value of theta.
But, you might ask, how can this be? After all, theta enters into the computation, and if you put in a different value for theta, you get different outputs. How could the gradient be zero?
What you must keep in mind, though, is that differentiation doesn't measure whether an input affects an output. It measures the change in output given an infinitesimal change in input.
Let's use a slightly simpler function as an example:
import jax
import jax.numpy as jnp
A = jnp.array([1.0, 2.0, 3.0])
theta = jnp.array([5.0, 1.0, 3.0])
def f(A, theta):
return A[jnp.argmax(theta)]
x = f(A, theta)
dx_dtheta = jax.grad(f, argnums=1)(A, theta)
Here the result of differentiating f with respect to theta is all zero, for the same reasons as above. Why? If you make an infinitesimal change to theta, it will in general not affect the sort order of theta. Thus, the entries you choose from A do not change given an infinitesimal change in theta, and thus the derivative with respect to theta is zero.
Now, you might argue that there are circumstances where this is not the case: for example, if two values in theta are very close together, then certainly perturbing one even infinitesimally could change their respective rank. This is true, but the gradient resulting from this procedure is undefined (the change in output is not smooth with respect to the change in input). The good news is this discontinuity is one-sided: if you perturb in the other direction, there is no change in rank and the gradient is well-defined. In order to avoid undefined gradients, most autodiff systems will implicitly use this safer definition of a derivative for rank-based computations.
The result is that the value of the output does not change when you infinitesimally perturb the input, which is another way of saying the gradient is zero. And this is not a failure of autodiff – it is the correct gradient given the definition of differentiation that autodiff is built on. Moreover, were you to try changing to a different definition of the derivative at these discontinuities, the best you could hope for would be undefined outputs, so the definition that results in zeros is arguably more useful and correct.
The numpy.cov(x, y) with 1-d array inputs returns the entire 2x2 covariance matrix. Is there a way to calculate only the cross-covariance, i.e. E[xy] - E[x]E[y] without wasting time on calculating the two variances?
PS. How does one write equations on here? The $$ does not seem to work as it does on other Stack websites.
What's wrong with simply implementing the equation you wrote?
# E[xy] - E[x]E[y]
cov = (x * y).mean() - x.mean() * y.mean()
I know that the Hessian matrix is a kind of second derivative test of functions involving more than one independent variable. How does one find the maximum or minimum of a function involving more than one variable? Is it found using the eigenvalues of the Hessian matrix or its principal minors?
You should have a look here:
https://en.wikipedia.org/wiki/Second_partial_derivative_test
For an n-dimensional function f, find an x where the gradient grad f = 0. This is a critical point.
Then, the 2nd derivatives tell, whether x marks a local minimum, a maximum or a saddle point.
The Hessian H is the matrix of all combinations of 2nd derivatives of f.
For the 2D-case the determinant and the minors of the Hessian are relevant.
For the nD-case it might involve a computation of eigen values of the Hessian H (if H is invertible) as part of checking H for being positive (or negative) definite.
In fact, the shortcut in 1) is generalized by 2)
For numeric calculations, some kind of optimization strategy can be used for finding x where grad f = 0.
I understand that chebvander2d and chebval2d return the Vandermonde matrix and fitted values for 2D inputs, and chebfit returns the coefficients for 1D-input series, but how do I get the coefficients for 2D-input series?
Short answer: It looks to me like this is not yet implemented. The whole of 2D polynomials seems more like a draft with some stub functions (as of June 2020).
Long answer (I came looking for the same thing, so I dug a little deeper):
First of all, this applies to all of the polynomial classes, not only chebyshev, so you also cannot fit an "ordinary" polynomial (power series). In fact, you cannot even construct one.
To understand the programming problem, let me recapture what a 2D polynomial looks like as a math formula, at an example polynomial of degree 2:
p(x, y) = c_00 + c_10 x + c_01 y + c_20 x^2 + c11 xy + c02 y^2
here the indices of c refer to the powers of x and y (the sum of the exponents must be <= degree).
First thing to notice is that, for degree d, there are (d+1)(d+2)/2 coefficients.
They could be stored in the upper left part of a matrix or in a 1D array, e.g. aranged as in the formula above.
The documentation of functions like numpy.polynomial.polynomial.polyval2d implies that numpy expects the matrix variant: p(x, y) = sum_i,j c_i,j * x^i * y^j.
Side note: it may be confusing that the row index i ("y-coordinate") of the matrix is used as exponent of x, not y; maybe the role of i and j should be switched if this is eventually implementd, or at least there should be a note in the documentation.
This leads to the core problem: the data structure for the 2D coefficients is not defined anywhere; only indirectly, like above, it can be guessed that a matrix should be used. But compared to a 1D array this is a waste of space, and evaluation of the polynomial takes two nested loops instead of just one. Also: does the matrix have to be initialized with np.zeros or do the implemented functions make sure that the lower right part is never touched so that np.empty can be used?
If the whole (d+1)^2 matrix were used, as the polyval2d function doc suggests, the degree of the polynomial would actually be d*2 (if c_d,d != 0)
To test this, I wanted to construct a numpy.polynomial.polynomial.Polynomial (yes, three times polynomial) and check the degree attribute:
import numpy as np
import numpy.polynomial.polynomial as poly
coef = np.array([
[5.00, 5.01, 5.02],
[5.10, 5.11, 0. ],
[5.20, 0. , 0. ]
])
polyObj = poly.Polynomial(coef)
print(polyObj.degree)
This gave a ValueError: Coefficient array is not 1-d before the print statement was reached. So while polyval2d expects a 2D coefficient array, it is not (yet) possible to construct such a polynomial - not manually like this at least. With this insight, it is not surprising that there is no function (yet) that computes a fit for 2D polynomials.
Given two matrices, A and B, where B is symetric (and positive semi-definite), What is the best (fastest) way to calculate A`*B*A?
Currently, using BLAS, I first compute C=B*A using dsymm (introducing a temporary matrix C) and then A`*C using dgemm.
Is there a better (faster, no temporaries) way to do this using BLAS and mkl?
Thanks.
I'll offer somekind of answer: Compared to the general case A*B*C you know that the end result is symmetric matrix. After computing C=B*A with BLAS subroutine dsymm, you want to compute A'C, but you only need to compute the upper diagonal part of the matrix and the copy the strictly upper diagonal part to the lower diagonal part.
Unfortunately there doesn't seem to be a BLAS routine where you can claim beforehand that given two general matrices, the output matrix will be symmetric. I'm not sure if it would be beneficial to write you own function for this. This probably depends on the size of your matrices and the implementation.
EDIT:
This idea seems to be addressed recently here: A Matrix Multiplication Routine that Updates Only the Upper or Lower Triangular Part of the Result Matrix