[Tests] Build fails with Numba 0.49.0

[QBatista]

See https://travis-ci.org/github/QuantEcon/QuantEcon.py/jobs/676806846

There are several issues:

• numba.targets.linalg was moved to numba.np.linalg (numba/numba@b0c408c)
• Tests for nelder_mead fail. This appears to be a bug with Numba because making changes in the following fashion at several places:

x_r = x_bar + ρ * (x_bar - vertices[worst_val_idx])

to

temp = ρ * (x_bar - vertices[worst_val_idx])
x_r = x_bar + temp

fixes the problem.

• Tests for quad fail. This appears to be related to the issue above because making the two following changes solves the issue:

pp = n * (z * p1 - p2)/(z * z - 1.0)

to

bottom = (z ** 2 - 1 )
pp = n * (z * p1 - p2) / bottom

and

weights[i] = 2 * xl / ((1 - z * z) * pp * pp)

to

weights[i] = 2 * xl / ((1 - z ** 2) * pp ** 2)

[oyamad]

@QBatista Thanks for your report.

I made a fix for the first problem #531.
For the second and third:

• Will you push your fixes to the numba049 branch to add to FIX: Updates for Numba 0.49.0 #531?
• Will you try to cook up a minimal working reproduce that reproduces the bug of Numba 0.49.0 as you discovered and then report it at the Numba issue tracker?

[QBatista]

@oyamad Thanks for this fix! Sure, I'll prepare the two fixes and try to cook up a minimal working example.

[QBatista]

@oyamad I've opened an issue on the Numba issue tracker: numba/numba#5599. For the other two fixes, it seems that I don't have permission to push directly to the numba049 branch.

[oyamad]

@QBatista Great work!

Will you push your fixes to your (or my) fork of QuantEcon.py? Then I will cherry-pick them into this PR. Thanks.

[QBatista]

@oyamad Sure -- I've pushed them to my fork here: https://github.com/QBatista/QuantEcon.py/commits/numba049.

[oyamad]

Thanks, they have been incorporated in the PR.

[stuartarchibald]

Thanks for reporting these problems on the Numba issue tracker, much appreciated.