I have two vectors w1 and w2 (each of length 100), and I want to minimize the sum of their absolute difference i.e.
import numpy as np
def diff(w: np.ndarray) -> float:
"""Get the sum of absolute differences in the vector w.
Args:
w: A flattened vector of length 200, with the first 100 elements
pertaining to w1, and the last 100 elements pertaining to w2.
Returns:
sum of absolute differences.
"""
return np.sum(np.absolute(w[:100] - w[-100:]))
I need to write diff() as only taking one argument since scipy.opyimize.minimize requires the array passed to the x0 argument to be 1 dimensional.
As for constraints, I have
w1 is fixed and not allowed to change
w2 is allowed to change
The sum of absolute values w2 is between 0.1 and 1.1: 0.1 <= sum(abs(w2)) <= 1.1
|w2_i| < 0.01 for any element i in w2
I am confused as to how we code these constraints using the Bounds and LinearConstraints objects. What I've tried so far is the following
from scipy.optimize import minimize, Bounds, LinearConstraint
bounds = Bounds(lb=[-0.01] * 200, ub=[0.01] * 200) # constraint #4
lc = LinearConstraint([[1] * 200], [0.1], [1.1]) # constraint #3
res = minimize(
fun=diff,
method='trust-constr',
x0=w, # my flattened vector containing w1 first 100 elements, and w2 in last 100 elements
bounds=bounds,
constraints=(lc)
)
My logic for the bounds variable is from constrain #4, and for the lc variable comes from constrain #3. However I know I've coded this wrong because because the lower and upper bounds are of length 200 which seems to indicate they are applied to both w1 and w2 whereas I only wan't to apply the constrains to w2 (I get an error ValueError: operands could not be broadcast together with shapes (200,) (100,) if I try to change the length of the array in Bounds from 200 to 100).
The shapes and argument types for LinearConstraint are especially confusing to me, but I did try to follow the scipy example.
This current implementation never seems to finish, it just hangs forever.
How do I properly implement bounds and LinearConstraint so that it satisfies my constraints list above, if that is even possible?
Your problem can easily be formulated as a linear optimization problem (LP). You only need to reformulate all absolute values of the optimization variables.
Changing the notation slightly (x is now the optimization variable w2 and w is just your given vector w1), your problem reads as
min |w_1 - x_1| + .... + |w_N - x_N|
s.t. lb <= |x1| + ... + |xN| <= ub (3)
|x_i| <= 0.01 - eps (4) (models the strict inequality)
where eps is just a sufficiently small number in order to model the strict inequality.
Let's consider the constraint (3). Here, we add additional positive variables z and define z_i = |x_i|. Then, we replace all absolute values |x_i| by z_i and impose the constraints -x_i <= z_i <= x_i which model the relationship z_i = |x_i|. Similarly, you can proceed with the objective and the constraint (4). The latter is by the way trivial and equivalent to -(0.01 - eps) <= x_i <= 0.01 - eps.
In the end, your optimization problem should read (assuming that all your w_i are positive):
min u1 + .... + uN
s.t. lb <= z1 + ... + zN <= ub
-x <= z <= x
-0.01 + eps <= x <= 0.01 - eps
-(w-x) <= u <= w - x
0 <= z
0 <= u
with 3*N optimization variables x1, ..., xN, u1, ..., uN, z1, ..., zN. It isn't hard to write these constraints as an matrix-vector product A_ineq * x <= b_ineq. Then, you can solve it by scipy.optimize.linprog as follows:
import numpy as np
from scipy.optimize import minimize, linprog
n = 100
w = np.abs(np.random.randn(n))
eps = 1e-10
lb = 0.1
ub = 1.1
# linear constraints: A_ub * (x, z, u)^T <= b_ub
A_ineq = np.block([
[np.zeros(n), np.ones(n), np.zeros(n)],
[np.zeros(n), -np.ones(n), np.zeros(n)],
[-np.eye(n), np.eye(n), np.zeros((n, n))],
[-np.eye(n), -np.eye(n), np.zeros((n, n))],
[ np.eye(n), np.zeros((n, n)), -np.eye(n)],
[ np.eye(n), np.zeros((n, n)), np.eye(n)],
])
b_ineq = np.hstack((ub, -lb, np.zeros(n), np.zeros(n), w, w))
# bounds: lower <= (x, z, u)^T <= upper
lower = np.hstack(((-0.01 + eps) * np.ones(n), np.zeros(n), np.zeros(n)))
upper = np.hstack((( 0.01 - eps) * np.ones(n), np.inf*np.ones(n), np.inf*np.ones(n)))
bounds = [(l, u) for (l, u) in zip(lower, upper)]
# objective: c^T * (x, z, u)
c = np.hstack((np.zeros(n), np.zeros(n), np.ones(n)))
# solve the problem
res = linprog(c, A_ub=A_ineq, b_ub=b_ineq, method="highs")
# your solution
x = res.x[:n]
print(res.message)
print(x)
Some notes in arbitrary order:
It's highly recommended to solve linear optimization problems with linprog instead of minimize. The former provides an interface to HiGHS, a high-performance LP solver HiGHs that outperforms all algorithms under the hood of minimize. However, it's also worth mentioning that minimize is meant to be used for nonlinear optimization problems.
In case your values w are not all positive, we need to change the formulation.
You can (and perhaps should, for clarity), use the args argument in minimize, and provide the fixed vector as an extra argument to your function.
If you set up your equation as follows:
def diff(w2, w1):
return np.sum(np.absolute(w1 - w2))
and your constraints with
bounds = Bounds(lb=[-0.01] * 100, ub=[0.01] * 100) # constraint #4
lc = LinearConstraint([[1] * 100], [0.1], [1.1]) # constraint #3
and then do
res = minimize(
fun=diff,
method='trust-constr',
x0=w1,
args=(w2,),
bounds=bounds,
constraints=[lc]
)
Then:
print(res.success, res.status, res.nit, np.abs(res.x).sum(), all(np.abs(res.x) < 0.01))
yields (for me at least)
(True, 1, 17, 0.9841520351691752, True)
which seems to be what you want.
Note that my test inputs are:
w1 = (np.arange(100) - 50) / 1000
w2 = np.ones(100, dtype=float)
which may or may not be favourable to the fitting procedure.
I do have a function, for example , but this can be something else as well, like a quadratic or logarithmic function. I am only interested in the domain of . The parameters of the function (a and k in this case) are known as well.
My goal is to fit a continuous piece-wise function to this, which contains alternating segments of linear functions (i.e. sloped straight segments, each with intercept of 0) and constants (i.e. horizontal segments joining the sloped segments together). The first and last segments are both sloped. And the number of segments should be pre-selected between around 9-29 (that is 5-15 linear steps + 4-14 constant plateaus).
Formally
The input function:
The fitted piecewise function:
I am looking for the optimal resulting parameters (c,r,b) (in terms of least squares) if the segment numbers (n) are specified beforehand.
The resulting constants (c) and the breakpoints (r) should be whole natural numbers, and the slopes (b) round two decimal point values.
I have tried to do the fitting numerically using the pwlf package using a segmented constant models, and further processed the resulting constant model with some graphical intuition to "slice" the constant steps with the slopes. It works to some extent, but I am sure this is suboptimal from both fitting perspective and computational efficiency. It takes multiple minutes to generate a fitting with 8 slopes on the range of 1-50000. I am sure there must be a better way to do this.
My idea would be to instead using only numerical methods/ML, the fact that we have the algebraic form of the input function could be exploited in some way to at least to use algebraic transforms (integrals) to get to a simpler optimization problem.
import numpy as np
import matplotlib.pyplot as plt
import pwlf
# The input function
def input_func(x,k,a):
return np.power(x,1/a)*k
x = np.arange(1,5e4)
y = input_func(x, 1.8, 1.3)
plt.plot(x,y);
def pw_fit(func, x_r, no_seg, *fparams):
# working on the specified range
x = np.arange(1,x_r)
y_input = func(x, *fparams)
my_pwlf = pwlf.PiecewiseLinFit(x, y_input, degree=0)
res = my_pwlf.fit(no_seg)
yHat = my_pwlf.predict(x)
# Function values at the breakpoints
y_isec = func(res, *fparams)
# Slope values at the breakpoints
slopes = np.round(y_isec / res, decimals=2)
slopes = slopes[1:]
# For the first slope value, I use the intersection of the first constant plateau and the input function
slopes = np.insert(slopes,0,np.round(y_input[np.argwhere(np.diff(np.sign(y_input - yHat))).flatten()[0]] / np.argwhere(np.diff(np.sign(y_input - yHat))).flatten()[0], decimals=2))
plateaus = np.unique(np.round(yHat))
# If due to rounding slope values (to two decimals), there is no change in a subsequent step, I just remove those segments
to_del = np.argwhere(np.diff(slopes) == 0).flatten()
slopes = np.delete(slopes,to_del + 1)
plateaus = np.delete(plateaus,to_del)
breakpoints = [np.ceil(plateaus[0]/slopes[0])]
for idx, j in enumerate(slopes[1:-1]):
breakpoints.append(np.floor(plateaus[idx]/j))
breakpoints.append(np.ceil(plateaus[idx+1]/j))
breakpoints.append(np.floor(plateaus[-1]/slopes[-1]))
return slopes, plateaus, breakpoints
slo, plat, breaks = pw_fit(input_func, 50000, 8, 1.8, 1.3)
# The piecewise function itself
def pw_calc(x, slopes, plateaus, breaks):
x = x.astype('float')
cond_list = [x < breaks[0]]
for idx, j in enumerate(breaks[:-1]):
cond_list.append((j <= x) & (x < breaks[idx+1]))
cond_list.append(breaks[-1] <= x)
func_list = [lambda x: x * slopes[0]]
for idx, j in enumerate(slopes[1:]):
func_list.append(plateaus[idx])
func_list.append(lambda x, j=j: x * j)
return np.piecewise(x, cond_list, func_list)
y_output = pw_calc(x, slo, plat, breaks)
plt.plot(x,y,y_output);
(Not important, but I think the fitted piecewise function is not continuous as it is. Intervals should be x<=r1; r1<x<=r2; ....)
As Anatolyg has pointed out, it looks to me that in the optimal solution (for the function posted at least, and probably for any where the derivative is different from zero), the horizantal segments will collapse to a point or the minimum segment length (in this case 1).
EDIT---------------------------------------------
The behavior above could only be valid if the slopes could have an intercept. If the intercepts are zero, as posted in the question, one consideration must be taken into account: Is the initial parabolic function defined in zero or nearby? Imagine the function y=0.001 *sqrt(x-1000), then the segments defined as b*x will have a slope close to zero and will be so similar to the constant segments that the best fit will be just the line that without intercept that fits better all the function.
Provided that the function is defined in zero or nearby, you can start by approximating the curve just by linear segments (with intercepts):
divide the function domain in N intervals(equal intervals or whose size is a function of the average curvature (or second derivative) of the function along the domain).
linear fit/regression in each intervals
for each interval, if a point (or bunch of points) in the extreme of any interval is better fitted by the line of the neighbor interval than the line of its interval, this point is assigned to the neighbor interval.
Repeat from 2) until no extreme points are moved.
Linear regressions might be optimized not to calculate all the covariance matrixes from scratch on each iteration, but just adding the contributions of the moved points to the previous covariance matrixes.
Then each linear segment (LSi) is replaced by a combination of a small constant segment at the beginning (Cbi), a linear segment without intercept (Si), and another constant segment at the end (Cei). This segments are easy to calculate as Si will contain the middle point of LSi, and Cbi and Cei will have respectively the begin and end values of the segment LSi. Then the intervals of each segment has to be calculated as an intersection between lines.
With this, the constant end segment will be collinear with the constant begin segment from the next interval so they will merge, resulting in a series of constant and linear segments interleaved.
But this would be a floating point start solution. Next, you will have to apply all the roundings which will mess up quite a lot all the segments as the conditions integer intervals and linear segments without slope can be very confronting. In fact, b,c,r are not totally independent. If ci and ri+1 are known, then bi+1 is already fixed
If nothing is broken so far, the final task will be to minimize the error/cost function (I assume that it will be the integral of the error between the parabolic function and the segments). My guess is that gradients here will be quite a pain, as if you change for example one ci, all the rest of the bj and cj will have to adapt as well due to the integer intervals restriction. However, if you can generalize the derivatives between parameters ( how much do I have to adapt bi+1 if ci changes a unit), you can propagate the change of one parameter to all other parameters and have kind of a gradient. Then for each interval, you can estimate what would be the ideal parameter and averaging all intervals calculate the best gradient step. Let me illustrate this:
Assuming first that r parameters are fixed, if I change c1 by one unit, b2 changes by 0.1, c2 changes by -0.2 and b3 changes by 0.2. This would be the gradient.
Then I estimate, comparing with the parabolic curve, that c1 should increase 0.5 (to reduce the cost by 10 points), b2 should increase 0.2 (to reduce the cost by 5 points), c2 should increase 0.2 (to reduce the cost by 6 points) and b3 should increase 0.1 (to reduce the cost by 9 points).
Finally, the gradient step would be (0.5/1·10 + 0.2/0.1·5 - 0.2/(-0.2)·6 + 0.1/0.2·9)/(10 + 5 + 6 + 9)~= 0.45. Thus, c1 would increase 0.45 units, b2 would increase 0.45·0.1, and so on.
When you add the r parameters to the pot, as integer intervals do not have an proper derivative, calculation is not straightforward. However, you can consider r parameters as floating points, calculate and apply the gradient step and then apply the roundings.
We can integrate the squared error function for linear and constant pieces and let SciPy optimize it. Python 3:
import matplotlib.pyplot as plt
import numpy as np
import scipy.optimize
xl = 1
xh = 50000
a = 1.3
p = 1 / a
n = 8
def split_b_and_c(bc):
return bc[::2], bc[1::2]
def solve_for_r(b, c):
r = np.empty(2 * n)
r[0] = xl
r[1:-1:2] = c / b[:-1]
r[2::2] = c / b[1:]
r[-1] = xh
return r
def linear_residual_integral(b, x):
return (
(x ** (2 * p + 1)) / (2 * p + 1)
- 2 * b * x ** (p + 2) / (p + 2)
+ b ** 2 * x ** 3 / 3
)
def constant_residual_integral(c, x):
return x ** (2 * p + 1) / (2 * p + 1) - 2 * c * x ** (p + 1) / (p + 1) + c ** 2 * x
def squared_error(bc):
b, c = split_b_and_c(bc)
r = solve_for_r(b, c)
linear = np.sum(
linear_residual_integral(b, r[1::2]) - linear_residual_integral(b, r[::2])
)
constant = np.sum(
constant_residual_integral(c, r[2::2])
- constant_residual_integral(c, r[1:-1:2])
)
return linear + constant
def evaluate(x, b, c, r):
i = 0
while x > r[i + 1]:
i += 1
return b[i // 2] * x if i % 2 == 0 else c[i // 2]
def main():
bc0 = (xl + (xh - xl) * np.arange(1, 4 * n - 2, 2) / (4 * n - 2)) ** (
p - 1 + np.arange(2 * n - 1) % 2
)
bc = scipy.optimize.minimize(
squared_error, bc0, bounds=[(1e-06, None) for i in range(2 * n - 1)]
).x
b, c = split_b_and_c(bc)
r = solve_for_r(b, c)
X = np.linspace(xl, xh, 1000)
Y = [evaluate(x, b, c, r) for x in X]
plt.plot(X, X ** p)
plt.plot(X, Y)
plt.show()
if __name__ == "__main__":
main()
I have tried to come up with a new solution myself, based on the idea of #Amo Robb, where I have partitioned the domain, and curve fitted a dual - constant and linear - piece together (with the help of np.maximum). I have used the 1 / f(x)' as the function to designate the breakpoints, but I know this is arbitrary and does not provide a global optimum. Maybe there is some optimal function for these breakpoints. But this solution is OK for me, as it might be appropriate to have a better fit at the first segments, at the expense of the error for the later segments. (The task itself is actually a cost based retail margin calculation {supply price -> added margin}, as the retail POS software can only work with such piecewise margin function).
The answer from #David Eisenstat is correct optimal solution if the parameters are allowed to be floats. Unfortunately the POS software can not use floats. It is OK to round up c-s and r-s afterwards. But the b-s should be rounded to two decimals, as those are inputted as percents, and this constraint would ruin the optimal solution with long floats. I will try to further improve my solution with both Amo's and David's valuable input. Thank You for that!
import numpy as np
import matplotlib.pyplot as plt
from scipy.optimize import curve_fit
# The input function f(x)
def input_func(x,k,a):
return np.power(x,1/a) * k
# 1 / f(x)'
def one_per_der(x,k,a):
return a / (k * np.power(x, 1/a-1))
# 1 / f(x)' inverted
def one_per_der_inv(x,k,a):
return np.power(a / (x*k), a / (1-a))
def segment_fit(start,end,y,first_val):
b, _ = curve_fit(lambda x,b: np.maximum(first_val, b*x), np.arange(start,end), y[start-1:end-1])
b = float(np.round(b, decimals=2))
bp = np.round(first_val / b)
last_val = np.round(b * end)
return b, bp, last_val
def pw_fit(end_range, no_seg, **fparams):
y_bps = np.linspace(one_per_der(1, **fparams), one_per_der(end_range,**fparams) , no_seg+1)[1:]
x_bps = np.round(one_per_der_inv(y_bps, **fparams))
y = input_func(x, **fparams)
slopes = [np.round(float(curve_fit(lambda x,b: x * b, np.arange(1,x_bps[0]), y[:int(x_bps[0])-1])[0]), decimals = 2)]
plats = [np.round(x_bps[0] * slopes[0])]
bps = []
for i, xbp in enumerate(x_bps[1:]):
b, bp, last_val = segment_fit(int(x_bps[i]+1), int(xbp), y, plats[i])
slopes.append(b); bps.append(bp); plats.append(last_val)
breaks = sorted(list(x_bps) + bps)[:-1]
# If due to rounding slope values (to two decimals), there is no change in a subsequent step, I just remove those segments
to_del = np.argwhere(np.diff(slopes) == 0).flatten()
breaks_to_del = np.concatenate((to_del * 2, to_del * 2 + 1))
slopes = np.delete(slopes,to_del + 1)
plats = np.delete(plats[:-1],to_del)
breaks = np.delete(breaks,breaks_to_del)
return slopes, plats, breaks
def pw_calc(x, slopes, plateaus, breaks):
x = x.astype('float')
cond_list = [x < breaks[0]]
for idx, j in enumerate(breaks[:-1]):
cond_list.append((j <= x) & (x < breaks[idx+1]))
cond_list.append(breaks[-1] <= x)
func_list = [lambda x: x * slopes[0]]
for idx, j in enumerate(slopes[1:]):
func_list.append(plateaus[idx])
func_list.append(lambda x, j=j: x * j)
return np.piecewise(x, cond_list, func_list)
fparams = {'k':1.8, 'a':1.2}
end_range = 5e4
no_steps = 10
x = np.arange(1, end_range)
y = input_func(x, **fparams)
slopes, plats, breaks = pw_fit(end_range, no_steps, **fparams)
y_output = pw_calc(x, slopes, plats, breaks)
plt.plot(x,y_output,y);
I have the same problem as in this question but don't want to add only one but several constraints to the optimization problem.
So e.g. I want to maximize x1 + 5 * x2 with the constraints that the sum of x1 and x2 is smaller than 5 and x2 is smaller than 3 (needless to say that the actual problem is far more complicated and cannot just thrown into scipy.optimize.minimize as this one; it just serves to illustrate the problem...).
I can to an ugly hack like this:
from scipy.optimize import differential_evolution
import numpy as np
def simple_test(x, more_constraints):
# check wether all constraints evaluate to True
if all(map(eval, more_constraints)):
return -1 * (x[0] + 5 * x[1])
# if not all constraints evaluate to True, return a positive number
return 10
bounds = [(0., 5.), (0., 5.)]
additional_constraints = ['x[0] + x[1] <= 5.', 'x[1] <= 3']
result = differential_evolution(simple_test, bounds, args=(additional_constraints, ), tol=1e-6)
print(result.x, result.fun, sum(result.x))
This will print
[ 1.99999986 3. ] -16.9999998396 4.99999985882
as one would expect.
Is there a better/ more straightforward way to add several constraints than using the rather 'dangerous' eval?
An example is something like this::
additional_constraints = [lambda(x): x[0] + x[1] <= 5., lambda(x):x[1] <= 3]
def simple_test(x, more_constraints):
# check wether all constraints evaluate to True
if all(constraint(x) for constraint in more_constraints):
return -1 * (x[0] + 5 * x[1])
# if not all constraints evaluate to True, return a positive number
return 10
There is a proper solution to the problem described in the question, to enforce multiple nonlinear constraints with scipy.optimize.differential_evolution.
The proper way is by using the scipy.optimize.NonlinearConstraint function.
Here below I give a non-trivial example of optimizing the classic Rosenbrock function inside a region defined by the intersection of two circles.
import numpy as np
from scipy import optimize
# Rosenbrock function
def fun(x):
return 100*(x[1] - x[0]**2)**2 + (1 - x[0])**2
# Function defining the nonlinear constraints:
# 1) x^2 + (y - 3)^2 < 4
# 2) (x - 1)^2 + (y + 1)^2 < 13
def constr_fun(x):
r1 = x[0]**2 + (x[1] - 3)**2
r2 = (x[0] - 1)**2 + (x[1] + 1)**2
return r1, r2
# No lower limit on constr_fun
lb = [-np.inf, -np.inf]
# Upper limit on constr_fun
ub = [4, 13]
# Bounds are irrelevant for this problem, but are needed
# for differential_evolution to compute the starting points
bounds = [[-2.2, 1.5], [-0.5, 2.2]]
nlc = optimize.NonlinearConstraint(constr_fun, lb, ub)
sol = optimize.differential_evolution(fun, bounds, constraints=nlc)
# Accurate solution by Mathematica
true = [1.174907377273171, 1.381484428610871]
print(f"nfev = {sol.nfev}")
print(f"x = {sol.x}")
print(f"err = {sol.x - true}\n")
This prints the following with default parameters:
nfev = 636
x = [1.17490808 1.38148613]
err = [7.06260962e-07 1.70116282e-06]
Here is a visualization of the function (contours) and the feasible region defined by the nonlinear constraints (shading inside the green line). The constrained global minimum is indicated by the yellow dot, while the magenta one shows the unconstrained global minimum.
This constrained problem has an obvious local minimum at (x, y) ~ (-1.2, 1.4) on the boundary of the feasible region which will make local optimizers fail to converge to the global minimum for many starting locations. However, differential_evolution consistently finds the global minimum as expected.
Consider a simple problem:
max log(x)
subject to x >= 1e-4
To solve the problem with scipy.optimize.minimize:
import numpy as np
from scipy.optimize import minimize
from math import log
def func(x):
return log(x[0])
def func_deriv(x):
return np.array([1 / x[0]])
cons = ({'type': 'ineq',
'fun' : lambda x: x[0] - 1e-4,
'jac' : lambda x: np.array([1])})
minimize(func, [1.0], jac=func_deriv, constraints=cons, method='SLSQP')
The script encounters ValueError because log(x) is evaluated with negative x. It seems that the function value is evaluated even if the constraint is not satisfied.
I understand that using bounds in minimize() could avoid the problem, but this is just a simplification of my original problem. In my original problem, the constraint x >= 1e-4 cannot be represented easily as bounds of x, but rather of the form g(x) >= C, so bounds wouldn't help.
If we only care about the function value with x > ε, it is possible to define a safe function extending the domain.
Take the log function as an example. It is possible to extend log with another cubic function, while making the bridge point ε smooth:
safe_log(x) = log(x) if x > ε else a * (x - b)**3
To calculate a and b, we have to satisfy:
log(ε) = a * (ε - b)**3
1 / ε = 3 * a * (ε - b)**2
Hence the safe_log function:
eps = 1e-3
def safe_log(x):
if x > eps:
return log(x)
logeps = log(eps)
a = 1 / (3 * eps * (3 * logeps * eps)**2)
b = eps * (1 - 3 * logeps)
return a * (x - b)**3
And it looks like this: