Related
I am trying to invert an interpolated function using scipy's interpolate function. Let's say I create an interpolated function,
import scipy.interpolate as interpolate
interpolatedfunction = interpolated.interp1d(xvariable,data,kind='cubic')
Is there some function that can find x when I specify a:
interpolatedfunction(x) == a
In other words, "I want my interpolated function to equal a; what is the value of xvariable such that my function is equal to a?"
I appreciate I can do this with some numerical scheme, but is there a more straightforward method? What if the interpolated function is multivalued in xvariable?
There are dedicated methods for finding roots of cubic splines. The simplest to use is the .roots() method of InterpolatedUnivariateSpline object:
spl = InterpolatedUnivariateSpline(x, y)
roots = spl.roots()
This finds all of the roots instead of just one, as generic solvers (fsolve, brentq, newton, bisect, etc) do.
x = np.arange(20)
y = np.cos(np.arange(20))
spl = InterpolatedUnivariateSpline(x, y)
print(spl.roots())
outputs array([ 1.56669456, 4.71145244, 7.85321627, 10.99554642, 14.13792756, 17.28271674])
However, you want to equate the spline to some arbitrary number a, rather than 0. One option is to rebuild the spline (you can't just subtract a from it):
solutions = InterpolatedUnivariateSpline(x, y - a).roots()
Note that none of this will work with the function returned by interp1d; it does not have roots method. For that function, using generic methods like fsolve is an option, but you will only get one root at a time from it. In any case, why use interp1d for cubic splines when there are more powerful ways to do the same kind of interpolation?
Non-object-oriented way
Instead of rebuilding the spline after subtracting a from data, one can directly subtract a from spline coefficients. This requires us to drop down to non-object-oriented interpolation methods. Specifically, sproot takes in a tck tuple prepared by splrep, as follows:
tck = splrep(x, y, k=3, s=0)
tck_mod = (tck[0], tck[1] - a, tck[2])
solutions = sproot(tck_mod)
I'm not sure if messing with tck is worth the gain here, as it's possible that the bulk of computation time will be in root-finding anyway. But it's good to have alternatives.
After creating an interpolated function interp_fn, you can find the value of x where interp_fn(x) == a by the roots of the function
interp_fn2 = lambda x: interp_fn(x) - a
There are number of options to find the roots in scipy.optimize. For instance, to use Newton's method with the initial value starting at 10:
from scipy import optimize
optimize.newton(interp_fn2, 10)
Actual example
Create an interpolated function and then find the roots where fn(x) == 5
import numpy as np
from scipy import interpolate, optimize
x = np.arange(10)
y = 1 + 6*np.arange(10) - np.arange(10)**2
y2 = 5*np.ones_like(x)
plt.scatter(x,y)
plt.plot(x,y)
plt.plot(x,y2,'k-')
plt.show()
# create the interpolated function, and then the offset
# function used to find the roots
interp_fn = interpolate.interp1d(x, y, 'quadratic')
interp_fn2 = lambda x: interp_fn(x)-5
# to find the roots, we need to supply a starting value
# because there are more than 1 root in our range, we need
# to supply multiple starting values. They should be
# fairly close to the actual root
root1, root2 = optimize.newton(interp_fn2, 1), optimize.newton(interp_fn2, 5)
root1, root2
# returns:
(0.76393202250021064, 5.2360679774997898)
If your data are monotonic you might also try the following:
inversefunction = interpolated.interp1d(data, xvariable, kind='cubic')
Mentioning another option because I found this page in a google search and the other option works for my simple use case. Hopefully it'll be of use to someone.
If the function you're interpolating is very simple and always has a 1:1 relationship between y and x, then you can simply take your data, swap x and y when you pass it into interp1d, and then call the interpolation function in that direction.
Adapting code from https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.interp1d.html
import numpy as np
import matplotlib.pyplot as plt
from scipy import interpolate
x = np.arange(0, 10)
y = np.exp(-x/3.0)
f = interpolate.interp1d(x, y)
xnew = np.arange(0, 9, 0.1)
ynew = f(xnew)
plt.plot(x, y, 'o', xnew, ynew, '-')
plt.show()
When x and y have been swapped you can call swappedInterpolationFunction(a) to get the x value where that would occur.
f = interpolate.interp1d(y, x)
xnew = np.arange(np.exp(-9/3), np.exp(0), 0.01)
ynew = f(xnew)
plt.plot(y, x, 'o', xnew, ynew, '-')
plt.title("Inverted")
plt.show()
Of course, if the function ever has multiple x values for a given y value (like sine or a parabola) then this will not work because it will no longer be a 1:1 function from x to y, and the above answers are necessary. This is just a simplification in a limited use case.
Background:
I am trying to implement a function doing an inverse transform sampling. I use sympy for calculating CDF and getting its inverse function. While for some simple PDFs I get correct results, for a PDF which CDF's inverse function includes Lambert-W function, results are wrong.
Example:
Consider following example CDF:
import sympy as sym
y = sym.Symbol('y')
cdf = (-y - 1) * sym.exp(-y) + 1 # derived from `pdf = x * sym.exp(-x)`
sym.plot(cdf, (y, -1, 5))
Now calculating inverse of this function:
x = sym.Symbol('x')
inverse = sym.solve(sym.Eq(x, cdf), y)
print(inverse)
Output:
[-LambertW((x - 1)*exp(-1)) - 1]
This, in fact, is only a left branch of negative y's of a given CDF:
sym.plot(inverse[0], (x, -0.5, 1))
Question:
How can I get the right branch for positive y's of a given CDF?
What I tried:
Specifying x and y to be only positive:
x = sym.Symbol('x', positive=True)
y = sym.Symbol('y', positive=True)
This doesn't have any effect, even for the first CDF plot.
Making CDF a Piecewise function:
cdf = sym.Piecewise((0, y < 0),
((-y - 1) * sym.exp(-y) + 1, True))
Again no effect. Strange thing here is that on another computer plotting this function gave a proper graph with zero for negative y's, but solving for a positive y's branch doesn't work anywhere. (Different versions? I also had to specify adaptive=False to sympy.plot to make it work there.)
Using sympy.solveset instead of sympy.solve:
This just gives a useless ConditionSet(y, Eq(x*exp(y) + y - exp(y) + 1, 0), Complexes(S.Reals x S.Reals, False)) as a result. Apparently, solveset still doesn't know how to deal with LambertW functions. From the docs:
When cases which are not solved or can only be solved incompletely, a
ConditionSet is used and acts as an unevaluated solveset object. <...>
There are still a few things solveset can’t do, which the old solve
can, such as solving non linear multivariate & LambertW type
equations.
Is it a bug or am I missing something? Is there any workaround to get the desired result?
The inverse produced by sympy is almost correct. The problem lies in the fact that the LambertW function has multiple branches over the domain (-1/e, 0). By default, it uses the upper branch, however for your problem you require the lower branch. The lower branch can be accessed by passing in a second argument to LambertW with a value of -1.
inverse = -sym.LambertW((x - 1)*sym.exp(-1), -1) - 1
sym.plot(inverse, (x, 0, 0.999))
Gives
I have a function, I want to get its integral function, something like this:
That is, instead of getting a single integration value at point x, I need to get values at multiple points.
For example:
Let's say I want the range at (-20,20)
def f(x):
return x**2
x_vals = np.arange(-20, 21, 1)
y_vals =[integrate.nquad(f, [[0, x_val]]) for x_val in x_vals ]
plt.plot(x_vals, y_vals,'-', color = 'r')
The problem
In the example code I give above, for each point, the integration is done from scratch. In my real code, the f(x) is pretty complex, and it's a multiple integration, so the running time is simply too slow(Scipy: speed up integration when doing it for the whole surface?).
I'm wondering if there is any way of efficient generating the Phi(x), at a giving range.
My thoughs:
The integration value at point Phi(20) is calucation from Phi(19), and Phi(19) is from Phi(18) and so on. So when we get Phi(20), in reality we also get the series of (-20,-19,-18,-17 ... 18,19,20). Except that we didn't save the value.
So I'm thinking, is it possible to create save points for a integrate function, so when it passes a save point, the value would get saved and continues to the next point. Therefore, by a single process toward 20, we could also get the value at (-20,-19,-18,-17 ... 18,19,20)
One could implement the strategy you outlined by integrating only over the short intervals (between consecutive x-values) and then taking the cumulative sum of the results. Like this:
import numpy as np
import scipy.integrate as si
def f(x):
return x**2
x_vals = np.arange(-20, 21, 1)
pieces = [si.quad(f, x_vals[i], x_vals[i+1])[0] for i in range(len(x_vals)-1)]
y_vals = np.cumsum([0] + pieces)
Here pieces are the integrals over short intervals, which get summed to produce y-values. As written, this code outputs a function that is 0 at the beginning of the range of integration which is -20. One can, of course, subtract the y-value that corresponds to x=0 in order to have the same normalization as on your plot.
That said, the split-and-sum process is unnecessary. When you find an indefinite integral of f, you are really solving the differential equation F' = f. And SciPy has a built-in method for that, odeint. Just use it:
import numpy as np
import scipy.integrate as si
def f(x):
return x**2
x_vals = np.arange(-20, 21, 1)
y_vals = si.odeint(lambda y,x: f(x), 0, x_vals)
The output is essential identical to the first version (within tiny computational errors), with less code. The reason for using lambda y,x: f(x) is that the first argument of odeint must be a function taking two arguments, the right-hand side of the equation y' = f(y, x).
For the equivalent version of user3717023's answer using scipy's solve_ivp you need to keep in mind the different ordering of x and y in the function f (different from the odeint version).
Further, keep in mind that you can only compute the solution up to a constant. So you might want to shift the result according to some given condition. In the example here (with the function f(x)=x^2 as given by the OP), I shifted the numeric solution such that it goes through the origin, matching the simplest analytic solution F(x)=x^3/3.
import numpy as np
import matplotlib.pyplot as plt
from scipy.integrate import solve_ivp
def f(x):
return x**2
xs = np.linspace(-20, 20, 1001)
# This is the integration step:
sol = solve_ivp(lambda x, y: f(x), t_span=(xs[0], xs[-1]), y0=[0], t_eval=xs)
plt.plot(sol.t, sol.t**3/3, ls='-', c='C0', label="analytic: $F(x)=x^3/3$")
plt.plot(sol.t, sol.y[0], ls='--', c='C1', label="numeric solution")
plt.plot(sol.t, sol.y[0] - sol.y[0][sol.t.size//2], ls='-.', c='C3', label="shifted solution going through origin")
plt.legend()
In case you don't have an analytical version of the function f, but only xs and ys as data points, then you can use scipy's interp1d function to interpolate between the data points and pass on that interpolating function the same way as before:
from scipy.interpolate import interp1d
f = interp1d(xs, ys)
I try to generate a generic fit polynom using SciPy's curve_fitmethod. My current simplified code looks like the following:
import functools
import scipy.optimize
def __fit_polynom_order_6(self, data):
def func(x, c1=None, c2=None, c3=None, c4=None, c5=None, c6=None):
return c1*x + c2*x**2 + c3*x**3 + c4*x**4 + c5*x**5 + c6*x**6
x, y = data[:,0], data[:,1]
popt, pcov = scipy.optimize.curve_fit(func, x, y)
func_fit = functools.partial(func, c1=popt[0],c2=popt[1],c3=popt[2],c4=popt[3],c5=popt[4],c6=popt[5])
return func_fit
Now I want also to do fits with polynoms of order n and thus generate a generic function __fit_polynom_order_n(self, n, data) that generates the polynom automatically and does essentially the same thing as my function above but with arbitrary polynoms.
My attempts doing this all came to nothing. Can you help? Thanks in advance!
There is already a function for that, np.polyfit:
fit = np.polyfit(x, y, n)
On the other hand, your func does not have a constant term. Is that on purpose?
If you wish to write your own polyfit-type method, you might want to study the source code for np.polyfit. You'll see that the problem is set up as a linear matrix equation and solved with np.linalg.lstsq, rather than the more general-purpose scipy.optimize.curve_fit.
# set up least squares equation for powers of x
lhs = vander(x, order)
rhs = y
c, resids, rank, s = lstsq(lhs, rhs, rcond)
Useful reference:
np.vander -- aha, this can be used to evaluate the polynomial at x. If you want to eliminate the constant term, you'd have to chop off the right-most column returned by np.vander.
Suppose I have a function f(x) defined between a and b. This function can have many zeros, but also many asymptotes. I need to retrieve all the zeros of this function. What is the best way to do it?
Actually, my strategy is the following:
I evaluate my function on a given number of points
I detect whether there is a change of sign
I find the zero between the points that are changing sign
I verify if the zero found is really a zero, or if this is an asymptote
U = numpy.linspace(a, b, 100) # evaluate function at 100 different points
c = f(U)
s = numpy.sign(c)
for i in range(100-1):
if s[i] + s[i+1] == 0: # oposite signs
u = scipy.optimize.brentq(f, U[i], U[i+1])
z = f(u)
if numpy.isnan(z) or abs(z) > 1e-3:
continue
print('found zero at {}'.format(u))
This algorithm seems to work, except I see two potential problems:
It will not detect a zero that doesn't cross the x axis (for example, in a function like f(x) = x**2) However, I don't think it can occur with the function I'm evaluating.
If the discretization points are too far, there could be more that one zero between them, and the algorithm could fail finding them.
Do you have a better strategy (still efficient) to find all the zeros of a function?
I don't think it's important for the question, but for those who are curious, I'm dealing with characteristic equations of wave propagation in optical fiber. The function looks like (where V and ell are previously defined, and ell is an positive integer):
def f(u):
w = numpy.sqrt(V**2 - u**2)
jl = scipy.special.jn(ell, u)
jl1 = scipy.special.jnjn(ell-1, u)
kl = scipy.special.jnkn(ell, w)
kl1 = scipy.special.jnkn(ell-1, w)
return jl / (u*jl1) + kl / (w*kl1)
Why are you limited to numpy? Scipy has a package that does exactly what you want:
http://docs.scipy.org/doc/scipy/reference/optimize.nonlin.html
One lesson I've learned: numerical programming is hard, so don't do it :)
Anyway, if you're dead set on building the algorithm yourself, the doc page on scipy I linked (takes forever to load, btw) gives you a list of algorithms to start with. One method that I've used before is to discretize the function to the degree that is necessary for your problem. (That is, tune \delta x so that it is much smaller than the characteristic size in your problem.) This lets you look for features of the function (like changes in sign). AND, you can compute the derivative of a line segment (probably since kindergarten) pretty easily, so your discretized function has a well-defined first derivative. Because you've tuned the dx to be smaller than the characteristic size, you're guaranteed not to miss any features of the function that are important for your problem.
If you want to know what "characteristic size" means, look for some parameter of your function with units of length or 1/length. That is, for some function f(x), assume x has units of length and f has no units. Then look for the things that multiply x. For example, if you want to discretize cos(\pi x), the parameter that multiplies x (if x has units of length) must have units of 1/length. So the characteristic size of cos(\pi x) is 1/\pi. If you make your discretization much smaller than this, you won't have any issues. To be sure, this trick won't always work, so you may need to do some tinkering.
I found out it's relatively easy to implement your own root finder using the scipy.optimize.fsolve.
Idea: Find any zeroes from interval (start, stop) and stepsize step by calling the fsolve repeatedly with changing x0. Use relatively small stepsize to find all the roots.
Can only search for zeroes in one dimension (other dimensions must be fixed). If you have other needs, I would recommend using sympy for calculating the analytical solution.
Note: It may not always find all the zeroes, but I saw it giving relatively good results. I put the code also to a gist, which I will update if needed.
import numpy as np
import scipy
from scipy.optimize import fsolve
from matplotlib import pyplot as plt
# Defined below
r = RootFinder(1, 20, 0.01)
args = (90, 5)
roots = r.find(f, *args)
print("Roots: ", roots)
# plot results
u = np.linspace(1, 20, num=600)
fig, ax = plt.subplots()
ax.plot(u, f(u, *args))
ax.scatter(roots, f(np.array(roots), *args), color="r", s=10)
ax.grid(color="grey", ls="--", lw=0.5)
plt.show()
Example output:
Roots: [ 2.84599497 8.82720551 12.38857782 15.74736542 19.02545276]
zoom-in:
RootFinder definition
import numpy as np
import scipy
from scipy.optimize import fsolve
from matplotlib import pyplot as plt
class RootFinder:
def __init__(self, start, stop, step=0.01, root_dtype="float64", xtol=1e-9):
self.start = start
self.stop = stop
self.step = step
self.xtol = xtol
self.roots = np.array([], dtype=root_dtype)
def add_to_roots(self, x):
if (x < self.start) or (x > self.stop):
return # outside range
if any(abs(self.roots - x) < self.xtol):
return # root already found.
self.roots = np.append(self.roots, x)
def find(self, f, *args):
current = self.start
for x0 in np.arange(self.start, self.stop + self.step, self.step):
if x0 < current:
continue
x = self.find_root(f, x0, *args)
if x is None: # no root found.
continue
current = x
self.add_to_roots(x)
return self.roots
def find_root(self, f, x0, *args):
x, _, ier, _ = fsolve(f, x0=x0, args=args, full_output=True, xtol=self.xtol)
if ier == 1:
return x[0]
return None
Test function
The scipy.special.jnjn does not exist anymore, but I created similar test function for the case.
def f(u, V=90, ell=5):
w = np.sqrt(V ** 2 - u ** 2)
jl = scipy.special.jn(ell, u)
jl1 = scipy.special.yn(ell - 1, u)
kl = scipy.special.kn(ell, w)
kl1 = scipy.special.kn(ell - 1, w)
return jl / (u * jl1) + kl / (w * kl1)
The main problem I see with this is if you can actually find all roots --- as have already been mentioned in comments, this is not always possible. If you are sure that your function is not completely pathological (sin(1/x) was already mentioned), the next one is what's your tolerance to missing a root or several of them. Put differently, it's about to what length you are prepared to go to make sure you did not miss any --- to the best of my knowledge, there is no general method to isolate all the roots for you, so you'll have to do it yourself. What you show is a reasonable first step already. A couple of comments:
Brent's method is indeed a good choice here.
First of all, deal with the divergencies. Since in your function you have Bessels in the denominators, you can first solve for their roots -- better look them up in e.g., Abramovitch and Stegun (Mathworld link). This will be a better than using an ad hoc grid you're using.
What you can do, once you've found two roots or divergencies, x_1 and x_2, run the search again in the interval [x_1+epsilon, x_2-epsilon]. Continue until no more roots are found (Brent's method is guaranteed to converge to a root, provided there is one).
If you cannot enumerate all the divergencies, you might want to be a little more careful in verifying a candidate is indeed a divergency: given x don't just check that f(x) is large, check that, e.g. |f(x-epsilon/2)| > |f(x-epsilon)| for several values of epsilon (1e-8, 1e-9, 1e-10, something like that).
If you want to make sure you don't have roots which simply touch zero, look for the extrema of the function, and for each extremum, x_e, check the value of f(x_e).
I've also encountered this problem to solve equations like f(z)=0 where f was an holomorphic function. I wanted to be sure not to miss any zero and finally developed an algorithm which is based on the argument principle.
It helps to find the exact number of zeros lying in a complex domain. Once you know the number of zeros, it is easier to find them. There are however two concerns which must be taken into account :
Take care about multiplicity : when solving (z-1)^2 = 0, you'll get two zeros as z=1 is counting twice
If the function is meromorphic (thus contains poles), each pole reduce the number of zero and break the attempt to count them.