All,
I've just been starting to play around with the Julia language and am enjoying it quite a bit. At the end of the 3rd tutorial there's an interesting problem: genericize the quadratic formula such that it solves for the roots of any n-order polynomial equation.
This struck me as (a) an interesting programming problem and (b) an interesting Julia problem. Has anyone out there solved this one? For reference, here is the Julia code with a couple toy examples. Again, the idea is to make this generic for any n-order polynomial.
Cheers,
Aaron
function derivative(f)
return function(x)
# pick a small value for h
h = x == 0 ? sqrt(eps(Float64)) : sqrt(eps(Float64)) * x
# floating point arithmetic gymnastics
xph = x + h
dx = xph - x
# evaluate f at x + h
f1 = f(xph)
# evaluate f at x
f0 = f(x)
# divide the difference by h
return (f1 - f0) / dx
end
end
function quadratic(f)
f1 = derivative(f)
c = f(0.0)
b = f1(0.0)
a = f(1.0) - b - c
return (-b + sqrt(b^2 - 4a*c + 0im))/2a, (-b - sqrt(b^2 - 4a*c + 0im))/2a
end
quadratic((x) -> x^2 - x - 2)
quadratic((x) -> x^2 + 2)
The package PolynomialRoots.jl provides the function roots() to find all (real and complex) roots of polynomials of any order. The only mandatory argument is the array with coefficients of the polynomial in ascending order.
For example, in order to find the roots of
6x^5 + 5x^4 + 3x^2 + 2x + 1
after loading the package (using PolynomialRoots) you can use
julia> roots([1, 2, 3, 4, 5, 6])
5-element Array{Complex{Float64},1}:
0.294195-0.668367im
-0.670332+2.77556e-17im
0.294195+0.668367im
-0.375695-0.570175im
-0.375695+0.570175im
The package is a Julia implementation of the root-finding algorithm described in this paper: http://arxiv.org/abs/1203.1034
PolynomialRoots.jl has also support for arbitrary precision calculation. This is useful for solving equation that cannot be solved in double precision. For example
julia> r = roots([94906268.375, -189812534, 94906265.625]);
julia> (r[1], r[2])
(1.0000000144879793 - 0.0im,1.0000000144879788 + 0.0im)
gives the wrong result for the polynomial, instead passing the input array in arbitrary precision forces arbitrary precision calculations that provide the right answer (see https://en.wikipedia.org/wiki/Loss_of_significance):
julia> r = roots([BigFloat(94906268.375), BigFloat(-189812534), BigFloat(94906265.625)]);
julia> (Float64(r[1]), Float64(r[2]))
(1.0000000289759583,1.0)
There are no algebraic formulae for a general polynomials of degree five and above (infact there cant be see here). So theoretically, you could proceed using the same methodology for solutions to cubics and quartics, but even that would be a lot of hard work given very unwieldy formulae for roots of quartics. You could also use a CAS like SymPy to find out those formulae.
Related
I wanted to try using the LSMR algorithm so I generated some data and run least squares. How come the LSMR solution and the closed-form one are different from the true $\beta$ I have used to generate the data?
using Distributions: Normal
using IterativeSolvers: lsmr
# Settings
n = 500
k = 50
# Generate data
X = rand(Normal(0.0, 0.1), (n, k)) + rand(Normal(0.0, 0.2), (n, k))
β = randn(k)
y = (X * β) + rand(Normal(0.0, 0.1), n)
# Solutions
closed_form_solution = (X'X) \ (X'y)
lsmr_solution = lsmr(X, y)
# Check solutions
β ≈ closed_form_solution, β ≈ lsmr_solution # returns false, false
I just ran your code and visualised the values, this is what I got.
using Plots
scatter(β)
scatter!(closed_form_solution)
scatter!(lsmr_solution)
As you can see they're actually pretty close, so the algorithms work as expected.
Now they're not the same because you've added residuals and the algorithms don't know about that. They have to adapt to them. You can find only an approximation of the underlying model if one even exists and this is basically what machine learning tries to do.
If you generate y's like this
y = X * β
and compare again you'll get much closer results (≈ might still return false though because of floating-point errors).
I'm trying to find one of the roots of a nonlinear (roughly quartic) equation.
The equation always has four roots, a pair of them close to zero, a large positive, and a large negative root. I'd like to identify either of the near zero roots, but nlsolve, even with an initial guess very close to these roots, seems to always converge on the large positive or negative root.
A plot of the function essentially looks like a constant negative value, with a (very narrow) even-ordered pole near zero, and gradually rising to cross zero at the large positive and negative roots.
Is there any way I can limit the region searched by nlsolve, or do something to make it more sensitive to the presence of this pole in my function?
EDIT:
Here's some example code reproducing the problem:
using NLsolve
function f!(F,x)
x = x[1]
F[1] = -15000 + x^4 / (x+1e-5)^2
end
# nlsolve will find the root at -122
nlsolve(f!,[0.0])
As output, I get:
Results of Nonlinear Solver Algorithm
* Algorithm: Trust-region with dogleg and autoscaling
* Starting Point: [0.0]
* Zero: [-122.47447713915808]
* Inf-norm of residuals: 0.000000
* Iterations: 15
* Convergence: true
* |x - x'| < 0.0e+00: false
* |f(x)| < 1.0e-08: true
* Function Calls (f): 16
* Jacobian Calls (df/dx): 6
We can find the exact roots in this case by transforming the objective function into a polynomial:
using PolynomialRoots
roots([-1.5e-6,-0.3,-15000,0,1])
produces
4-element Array{Complex{Float64},1}:
122.47449713915809 - 0.0im
-122.47447713915808 + 0.0im
-1.0000000813048448e-5 + 0.0im
-9.999999186951818e-6 + 0.0im
I would love a way to identify the pair of roots around the pole at x = -1e-5 without knowing the exact form of the objective function.
EDIT2:
Trying out Roots.jl :
using Roots
f(x) = -15000 + x^4 / (x+1e-5)^2
find_zero(f,0.0) # finds +122... root
find_zero(f,(-1e-4,0.0)) # error, not a bracketing interval
find_zeros(f,-1e-4,0.0) # finds 0-element Array{Float64,1}
find_zeros(f,-1e-4,0.0,no_pts=6) # finds root slightly less than -1e-5
find_zeros(f,-1e-4,0.0,no_pts=10) # finds 0-element Array{Float64,1}, sensitive to value of no_pts
I can get find_zeros to work, but it's very sensitive to the no_pts argument and the exact values of the endpoints I pick. Doing a loop over no_pts and taking the first non-empty result might work, but something more deterministic to converge would be preferable.
EDIT3 :
Here's applying the tanh transformation suggested by Bogumił
using NLsolve
function f_tanh!(F,x)
x = x[1]
x = -1e-4 * (tanh(x)+1) / 2
F[1] = -15000 + x^4 / (x+1e-5)^2
end
nlsolve(f_tanh!,[100.0]) # doesn't converge
nlsolve(f_tanh!,[1e5]) # doesn't converge
using Roots
function f_tanh(x)
x = -1e-4 * (tanh(x)+1) / 2
return -15000 + x^4 / (x+1e-5)^2
end
find_zeros(f_tanh,-1e10,1e10) # 0-element Array
find_zeros(f_tanh,-1e3,1e3,no_pts=100) # 0-element Array
find_zero(f_tanh,0.0) # convergence failed
find_zero(f_tanh,0.0,max_evals=1_000_000,maxfnevals=1_000_000) # convergence failed
EDIT4 : This combination of techniques identifies at least one root somewhere around 95% of the time, which is good enough for me.
using Peaks
using Primes
using Roots
# randomize pole location
a = 1e-4*rand()
f(x) = -15000 + x^4 / (x+a)^2
# do an initial sample to find the pole location
l = 1000
minval = -1e-4
maxval = 0
m = []
sample_r = []
while l < 1e6
sample_r = range(minval,maxval,length=l)
rough_sample = f.(sample_r)
m = maxima(rough_sample)
if length(m) > 0
break
else
l *= 10
end
end
guess = sample_r[m[1]]
# functions to compress the range around the estimated pole
cube(x) = (x-guess)^3 + guess
uncube(x) = cbrt(x-guess) + guess
f_cube(x) = f(cube(x))
shift = l ÷ 1000
low = sample_r[m[1]-shift]
high = sample_r[m[1]+shift]
# search only over prime no_pts, so no samplings divide into each other
# possibly not necessary?
for i in primes(500)
z = find_zeros(f_cube,uncube(low),uncube(high),no_pts=i)
if length(z)>0
println(i)
println(cube.(z))
break
end
end
More comment could be given if you provided more information on your problem.
However in general:
It seems that your problem is univariate, in which case you can use Roots.jl where find_zero and find_zeros give the interface you ask for (i.e. allowing to specify the search region)
If a problem is multivariate you have several options how to do it in the problem specification for nlsolve (as it by default does not allow to specify a bounding box AFAICT). The simplest is to use variable transformation. E.g. you can apply a ai * tanh(xi) + bi transformation selecting ai and bi for each variable so that it is bounded to the desired interval
The first problem you have in your definition is that the way you define f it never crosses 0 near the two roots you are looking for because Float64 does not have enough precision when you write 1e-5. You need to use greater precision of computations:
julia> using Roots
julia> f(x) = -15000 + x^4 / (x+1/big(10.0^5))^2
f (generic function with 1 method)
julia> find_zeros(f,big(-2*10^-5), big(-8*10^-6), no_pts=100)
2-element Array{BigFloat,1}:
-1.000000081649671426108658262468117284940444265467160592853348997523986352593615e-05
-9.999999183503552405580084054429938261707450678661727461293670518591720605751116e-06
and set no_pts to be sufficiently large to find intervals bracketing the roots.
I'm using symbolic math in Julia. When I do the differentiation it comes out very nicely, but I cant get the coefficients out
using SymPy
#vars x y
z = x*(10 + 5*x + 4*y)
dz = diff(z,x)
x_s = solveset(dz,x)
how do I get the the coefficients out of x_s?
You can use elements to get the elements of a finite sets as an array:
julia> elements(s)
1-element Array{Sym,1}:
-2*y/5 - 1
To get the coefficients can be done different ways, but here we convert the value to a Polynomial type, then use its coeffs method:
julia> p = sympy.Poly(elements(s)[1], y)
Poly(-2/5*y - 1, y, domain='QQ')
julia> p.coeffs()
2-element Array{Sym,1}:
-2/5
-1
As per my comment, the following works but isn't exactly what I'd describe as pretty:
julia> x_s.__pyobject__.args[1].__pyobject__.args[1]
-1
julia> x_s.__pyobject__.args[1].__pyobject__.args[2]
-2⋅y
─────
5
julia> x_s.__pyobject__.args[1].__pyobject__.args[2].__pyobject__.args[1]
-2/5
I couldn't find an accessor function in Sympy.jl that simplifies this, but as you say this could be the basis for rolling your own.
EDIT 2: this post seems to have been moved from CrossValidated to StackOverflow due to it being mostly about programming, but that means by fancy MathJax doesn't work anymore. Hopefully this is still readable.
Say I want to to calculate the squared Mahalanobis distance between two vectors x and y with covariance matrix S. This is a fairly simple function defined by
M2(x, y; S) = (x - y)^T * S^-1 * (x - y)
With python's numpy package I can do this as
# x, y = numpy.ndarray of shape (n,)
# s_inv = numpy.ndarray of shape (n, n)
diff = x - y
d2 = diff.T.dot(s_inv).dot(diff)
or in R as
diff <- x - y
d2 <- t(diff) %*% s_inv %*% diff
In my case, though, I am given
m by n matrix X
n-dimensional vector mu
n by n covariance matrix S
and want to find the m-dimensional vector d such that
d_i = M2(x_i, mu; S) ( i = 1 .. m )
where x_i is the ith row of X.
This is not difficult to accomplish using a simple loop in python:
d = numpy.zeros((m,))
for i in range(m):
diff = x[i,:] - mu
d[i] = diff.T.dot(s_inv).dot(diff)
Of course, given that the outer loop is happening in python instead of in native code in the numpy library means it's not as fast as it could be. $n$ and $m$ are about 3-4 and several hundred thousand respectively and I'm doing this somewhat often in an interactive program so a speedup would be very useful.
Mathematically, the only way I've been able to formulate this using basic matrix operations is
d = diag( X' * S^-1 * X'^T )
where
x'_i = x_i - mu
which is simple to write a vectorized version of, but this is unfortunately outweighed by the inefficiency of calculating a 10-billion-plus element matrix and only taking the diagonal... I believe this operation should be easily expressible using Einstein notation, and thus could hopefully be evaluated quickly with numpy's einsum function, but I haven't even begun to figure out how that black magic works.
So, I would like to know: is there either a nicer way to formulate this operation mathematically (in terms of simple matrix operations), or could someone suggest some nice vectorized (python or R) code that does this efficiently?
BONUS QUESTION, for the brave
I don't actually want to do this once, I want to do it k ~ 100 times. Given:
m by n matrix X
k by n matrix U
Set of n by n covariance matrices each denoted S_j (j = 1..k)
Find the m by k matrix D such that
D_i,j = M(x_i, u_j; S_j)
Where i = 1..m, j = 1..k, x_i is the ith row of X and u_j is the jth row of U.
I.e., vectorize the following code:
# s_inv is (k x n x n) array containing "stacked" inverses
# of covariance matrices
d = numpy.zeros( (m, k) )
for j in range(k):
for i in range(m):
diff = x[i, :] - u[j, :]
d[i, j] = diff.T.dot(s_inv[j, :, :]).dot(diff)
First off, it seems like maybe you're getting S and then inverting it. You shouldn't do that; it's slow and numerically inaccurate. Instead, you should get the Cholesky factor L of S so that S = L L^T; then
M^2(x, y; L L^T)
= (x - y)^T (L L^T)^-1 (x - y)
= (x - y)^T L^-T L^-1 (x - y)
= || L^-1 (x - y) ||^2,
and since L is triangular L^-1 (x - y) can be computed efficiently.
As it turns out, scipy.linalg.solve_triangular will happily do a bunch of these at once if you reshape it properly:
L = np.linalg.cholesky(S)
y = scipy.linalg.solve_triangular(L, (X - mu[np.newaxis]).T, lower=True)
d = np.einsum('ij,ij->j', y, y)
Breaking that down a bit, y[i, j] is the ith component of L^-1 (X_j - \mu). The einsum call then does
d_j = \sum_i y_{ij} y_{ij}
= \sum_i y_{ij}^2
= || y_j ||^2,
like we need.
Unfortunately, solve_triangular won't vectorize across its first argument, so you should probably just loop there. If k is only about 100, that's not going to be a significant issue.
If you are actually given S^-1 rather than S, then you can indeed do this with einsum more directly. Since S is quite small in your case, it's also possible that actually inverting the matrix and then doing this would be faster. As soon as n is a nontrivial size, though, you're throwing away a lot of numerical accuracy by doing this.
To figure out what to do with einsum, write everything in terms of components. I'll go straight to the bonus case, writing S_j^-1 = T_j for notational convenience:
D_{ij} = M^2(x_i, u_j; S_j)
= (x_i - u_j)^T T_j (x_i - u_j)
= \sum_k (x_i - u_j)_k ( T_j (x_i - u_j) )_k
= \sum_k (x_i - u_j)_k \sum_l (T_j)_{k l} (x_i - u_j)_l
= \sum_{k l} (X_{i k} - U_{j k}) (T_j)_{k l} (X_{i l} - U_{j l})
So, if we make arrays X of shape (m, n), U of shape (k, n), and T of shape (k, n, n), then we can write this as
diff = X[np.newaxis, :, :] - U[:, np.newaxis, :]
D = np.einsum('jik,jkl,jil->ij', diff, T, diff)
where diff[j, i, k] = X_[i, k] - U[j, k].
Dougal nailed this one with an excellent and detailed answer, but thought I'd share a small modification that I found increases efficiency in case anyone else is trying to implement this. Straight to the point:
Dougal's method was as follows:
def mahalanobis2(X, mu, sigma):
L = np.linalg.cholesky(sigma)
y = scipy.linalg.solve_triangular(L, (X - mu[np.newaxis,:]).T, lower=True)
return np.einsum('ij,ij->j', y, y)
A mathematically equivalent variant I tried is
def mahalanobis2_2(X, mu, sigma):
# Cholesky decomposition of inverse of covariance matrix
# (Doing this in either order should be equivalent)
linv = np.linalg.cholesky(np.linalg.inv(sigma))
# Just do regular matrix multiplication with this matrix
y = (X - mu[np.newaxis,:]).dot(linv)
# Same as above, but note different index at end because the matrix
# y is transposed here compared to above
return np.einsum('ij,ij->i', y, y)
Ran both versions head-to-head 20x using identical random inputs and recorded the times (in milliseconds). For X as a 1,000,000 x 3 matrix (mu and sigma 3 and 3x3) I get:
Method 1 (min/max/avg): 30/62/49
Method 2 (min/max/avg): 30/47/37
That's about a 30% speedup for the 2nd version. I'm mostly going to be running this in 3 or 4 dimensions but to see how it scaled I tried X as 1,000,000 x 100 and got:
Method 1 (min/max/avg): 970/1134/1043
Method 2 (min/max/avg): 776/907/837
which is about the same improvement.
I mentioned this in a comment on Dougal's answer but adding here for additional visibility:
The first pair of methods above take a single center point mu and covariance matrix sigma and calculate the squared Mahalanobis distance to each row of X. My bonus question was to do this multiple times with many sets of mu and sigma and output a two-dimensional matrix. The set of methods above can be used to accomplish this with a simple for loop, but Dougal also posted a more clever example using einsum.
I decided to compare these methods with each other by using them to solve the following problem: Given k d-dimensional normal distributions (with centers stored in rows of k by d matrix U and covariance matrices in the last two dimensions of the k by d by d array S), find the density at the n points stored in rows of the n by d matrix X.
The density of a multivariate normal distribution is a function of the squared Mahalanobis distance of the point to the mean. Scipy has an implementation of this as scipy.stats.multivariate_normal.pdf to use as a reference. I ran all three methods against each other 10x using identical random parameters each time, with d=3, k=96, n=5e5. Here are the results, in points/sec:
[Method]: (min/max/avg)
Scipy: 1.18e5/1.29e5/1.22e5
Fancy 1: 1.41e5/1.53e5/1.48e5
Fancy 2: 8.69e4/9.73e4/9.03e4
Fancy 2 (cheating version): 8.61e4/9.88e4/9.04e4
where Fancy 1 is the better of the two methods above and Fancy2 is Dougal's 2nd solution. Since the Fancy 2 needs to calculate the inverses of all the covariance matrices I also tried a "cheating version" where it was passed these as a parameter, but it looks like that didn't make a difference. I had planned on including the non-vectorized implementation but that was so slow it would have taken all day.
What we can take away from this is that using Dougal's first method is about 20% faster than however Scipy does it. Unfortunately despite its cleverness the 2nd method is only about 60% as fast as the first. There are probably some other optimizations that can be done but this is already fast enough for me.
I also tested how this scaled with higher dimensionality. With d=100, k=96, n=1e4:
Scipy: 7.81e3/7.91e3/7.86e3
Fancy 1: 1.03e4/1.15e4/1.08e4
Fancy 2: 3.75e3/4.10e3/3.95e3
Fancy 2 (cheating version): 3.58e3/4.09e3/3.85e3
Fancy 1 seems to have an even bigger advantage this time. Also worth noting that Scipy threw a LinAlgError 8/10 times, probably because some of my randomly-generated 100x100 covariance matrices were close to singular (which may mean that the other two methods are not as numerically stable, I did not actually check the results).
For an ocean shader, I need a fast function that computes a very approximate value for sin(x). The only requirements are that it is periodic, and roughly resembles a sine wave.
The taylor series of sin is too slow, since I'd need to compute up to the 9th power of x just to get a full period.
Any suggestions?
EDIT: Sorry I didn't mention, I can't use a lookup table since this is on the vertex shader. A lookup table would involve a texture sample, which on the vertex shader is slower than the built in sin function.
It doesn't have to be in any way accurate, it just has to look nice.
Use a Chebyshev approximation for as many terms as you need. This is particularly easy if your input angles are constrained to be well behaved (-π .. +π or 0 .. 2π) so you do not have to reduce the argument to a sensible value first. You might use 2 or 3 terms instead of 9.
You can make a look-up table with sin values for some values and use linear interpolation between that values.
A rational algebraic function approximation to sin(x), valid from zero to π/2 is:
f = (C1 * x) / (C2 * x^2 + 1.)
with the constants:
c1 = 1.043406062
c2 = .2508691922
These constants were found by least-squares curve fitting. (Using subroutine DHFTI, by Lawson & Hanson).
If the input is outside [0, 2π], you'll need to take x mod 2 π.
To handle negative numbers, you'll need to write something like:
t = MOD(t, twopi)
IF (t < 0.) t = t + twopi
Then, to extend the range to 0 to 2π, reduce the input with something like:
IF (t < pi) THEN
IF (t < pi/2) THEN
x = t
ELSE
x = pi - t
END IF
ELSE
IF (t < 1.5 * pi) THEN
x = t - pi
ELSE
x = twopi - t
END IF
END IF
Then calculate:
f = (C1 * x) / (C2 * x*x + 1.0)
IF (t > pi) f = -f
The results should be within about 5% of the real sine.
Well, you don't say how accurate you need it to be. The sine can be approximated by straight lines of slopes 2/pi and -2/pi on intervals [0, pi/2], [pi/2, 3*pi/2], [3*pi/2, 2*pi]. This approximation can be had for the cost of a multiplication and an addition after reducing the angle mod 2*pi.
Using a lookup table is probably the best way to control the tradeoff between speed and accuracy.