New initialisation method when passin a df.Field using the new xarray functionality.

[lang-m]

Test

• field with mesh.n = (100, 100, 10)
• creation of a new field with mesh.n = (10, 10, 10) and passing the old field to value

Performance improvement:

• old implementation (using that the field is callable): ~ 6.5 s
• new implementation (interpolation done by xarray): ~ 5 ms

@swapneelap This is probably the first use case for your new method.

[swapneelap]

@lang-m Happy to see that it is useful to the project! :)
Great performance improvement BTW.

[marijanbeg]

Looks very good! Just added a few minor suggestions.

Is there a reason why we want all _as_array in Field class and not in dfu?

[lang-m]

@marijanbeg Could you please have a look at the last test failure. This is a problem with floating-point accuracy. My suggestion would be to slightly relax region.__contains__ by adding numpy.isclose tests for pmin and pmax. This would mean going from other.pmin >= self.pmin to other.pmin > or "approx equal" self.pmin. Have you got any other ideas?

[marijanbeg]

@lang-m, I agree. I am just not sure what is the right way of avoiding hard-coded tolerances. Maybe we could define some tolerances in utils and use them throughout df.

[lang-m]

@marijanbeg Do you think sticking to the default settings for rtol and atol of np.isclose wouldn't be sufficient? Mabye with a smaller value for atol, the default is 1e-8

[lang-m]

The new _as_array implementation depending on df.Field cannot be in dfu because we run into issues with circular imports. To make it consistent I decided to move everything to the field module as this also is the only place where we need theses functions.

[marijanbeg]

Nice, I like that.

[marijanbeg]

@lang-m, default value for rtol is fine. Only atol could be an issue because the values we have in the field can be in a wide range. Probably setting it to zero is fine.

[codecov-commenter]

Codecov Report

Merging #135 (451cfc2) into master (f831fa2) will decrease coverage by 0.04%.
The diff coverage is 93.65%.

@@            Coverage Diff             @@
##           master     #135      +/-   ##
==========================================
- Coverage   95.86%   95.82%   -0.05%     
==========================================
  Files          20       20              
  Lines        2106     2108       +2     
==========================================
+ Hits         2019     2020       +1     
- Misses         87       88       +1     

Impacted Files	Coverage Δ
discretisedfield/util/util.py	97.97% <ø> (+1.42%)	⬆️
discretisedfield/field.py	97.47% <92.45%> (-0.40%)	⬇️
discretisedfield/ovf2vtk.py	100.00% <100.00%> (ø)
discretisedfield/region.py	98.75% <100.00%> (+0.03%)	⬆️
discretisedfield/util/__init__.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f831fa2...451cfc2. Read the comment docs.

[lang-m]

@marijanbeg Setting atol to 0 does not work when one of the points is at (0, 0, 0) (or probably also just 0 in one direction, I didn't check that). I've now introduced a different way of calculating rtol and atol. Could you please review d789e1c.

[marijanbeg]

It does exactly what it says on the tin.