Wrong sign for right boundary with Sympy

[Davide-sd]

Let's consider the following simple example. First, I'm going to show the correct result when using Numpy:

import numpy as np
from adaptive import notebook_extension, Learner1D
from adaptive.runner import simple
import dill

notebook_extension()

def func(x):
    return np.imag(np.sqrt(-(x + 0j)))

def goal(learner):
    return learner.loss() < 0.1

learner = Learner1D(func, bounds=(-1, 1))
simple(learner, goal)
learner.plot()

Now, I'm going to use Sympy to generate a function which is going to be evaluated with Numpy:

from sympy import *
var("x")
f = im(sqrt(-x))

# func is going to use Numpy
func = lambdify(x, f)

def goal(learner):
    return learner.loss() < 0.1

learner = Learner1D(func, bounds=(-1, 1))
learner_bytes = dill.dumps(learner)
learner_loaded = dill.loads(learner_bytes)
simple(learner_loaded, goal)
learner_loaded.plot()

As you can see, the last evaluation point (the right boundary) has the wrong sign. Even if I change the boundaries, the last point will always have the wrong sign. Why does it happen? Is it something that I can fix with some option in the learner/runner or is it an internal bug?

[basnijholt]

Thanks for reporting!

That looks like a bug, I will try to look into it today or tomorrow.

[basnijholt]

Without using Adaptive:

from sympy import im, sqrt, lambdify, var
var("x")
f = im(sqrt(-x))
func = lambdify(x, f)
print(func(1))

prints: 1

and even:

from sympy import im, sqrt
im(sqrt(-1))

prints: 1

edit: however, this is the expected result, right?

[Davide-sd]

Interesting, Numpy deals with (complex number) branch cuts differently depending on the input data type. For example f_numpy(1), f_numpy(1.0) outputs (1.0, -1.0). It appears that there is no bug with adaptive. Thanks for pointing me in the right direction. All I need to do is to use float boundaries. I'm closing the issue.