Use np.interp instead of interpolation.interp

[kp992]

This PR replaces interpolation.interp to np.interp now that numba supports np.interp directly. Also, removes plt.rcParams["figure.figsize"] = (11, 5) #set default figure size redundant line.

[kp992]

@jstac @mmcky I am trying to replace interp function in this PR. There are a few more files which needs this change and will do that in new PR to keep this PR short and easy to review and verify. Thank you!

[github-actions]

🚀 Deployed on https://65ebef747310825342956403--nostalgic-wright-5fa355.netlify.app

[jstac]

Many thanks @kp992 .

@mmcky or @HumphreyYang , could you please do a first round review of this?

The aim is that swapping to NumPy's interp should not change outputs or degrade performance.

[jstac]

Note that this is related to QuantEcon/lecture-python-advanced.myst#156

[mmcky]

thanks @kp992 I confirm the np.interp has been updated nicely and the function arguments have been switched successfully.

[mmcky]

@jstac re: performance

lecture	current time	new time
cake_eating_numerical	30.4	33.8
coleman_policy_iter	24.58	21.52
egm_policy_iter	12.79	6.89
ifp	40.88	51.77
ifp_advanced	32.49	30.85

Some lectures actually improved and some are a bit slower. I don't see any major timing issues.

[mmcky]

Remaining lectures to migrate

• lectures/jv.md:from interpolation import interp
• lectures/mccall_correlated.md:from interpolation import interp
• lectures/mccall_fitted_vfi.md:from interpolation import interp
• lectures/navy_captain.md:from interpolation import interp
• lectures/odu.md:from interpolation import mlinterp, interp
• lectures/optgrowth_fast.md:from interpolation import interp
• lectures/wald_friedman.md:from interpolation import interp

@kp992 can you let me know if there is an equivalent to mlinterp in numpy that is compatible with numba

[mmcky]

@kp992 I know it violates our usual practice but let's try and fix them all in one PR -- so they are in one commit. This lecture series is not in high frequency editing at the moment so I am happy with that approach.

[mmcky]

@kp992 is there a corresponding function in numpy for mlinterp?

[jstac]

No @mmcky, there's not.

There is in SciPy but it's not jit compilable. Let's stick to interpolation for the multivariate case.

[kp992]

That's right, @jstac. I didn't find mlinterp in numpy too.

[mmcky]

@kp992 we will need interpolation package updated unless mlinterp doesn't use generated_jit. Would you have time to look into that?

[mmcky]

thanks @kp992.

@jstac I think this is looking good -- and I have reviewed the function calls. @kp992 has done a nice job.

[jstac]

I didn't see pip install interpolation removed from this lecture... was that task forgotten?

[kp992]

Thanks @jstac. I think we can remove this file completely. This isn't use anywhere in any lectures. In optgrowth, it seems to be have been replace by from scipy.interpolate import interp1d.

% git grep bellman_operator
lectures/_static/lecture_specific/odu/odu.py:    def bellman_operator(self, v):
lectures/_static/lecture_specific/wald_friedman/wald_class.py:    def bellman_operator(self, J):
lectures/_static/lecture_specific/wald_friedman/wald_class.py:        J = qe.compute_fixed_point(self.bellman_operator, np.zeros(self.m),
lectures/_static/lecture_specific/wald_friedman/wf_first_pass.py:def bellman_operator(pgrid, c, f0, f1, L0, L1, J):
lectures/_static/lecture_specific/wald_friedman/wf_first_pass.py:bell_op = lambda vf: bellman_operator(pg, 0.5, f0, f1, 5.0, 5.0, vf)

[jstac]

Thanks @mmcky and great work @kp992 .

Please see minor comments above.

[mmcky]

thanks @jstac for searching the py includes. I had focused on the md 👍

[mmcky]

@kp992 are you happy for this to be merged?

[kp992]

Thanks @mmcky and @jstac