[feature] Simpler way to add derived fields

[cphyc]

PR Summary

This adds a simple syntax to define a new derived field.
This allows to write:

ds = yt.load(...)

ds.fields.gas.density_square = ds.fields.gas.density * ds.fields.gas.density
ds.add_field(("gas", "double_density"), ds.fields.gas.density * 2)

ds.r["gas", "density_square"]
ds.r["gas", "double_density"]

Open for discussion before I write the documentation (or we kill that PR :) )

PR Checklist

• New features are documented, with docstrings and narrative docs
• Adds a test for any bugs fixed. Adds tests for new features.

[matthewturk]

I genuinely love this and I find it to be an incredibly nice, ergonomic way to add fields.

[matthewturk]

I still think this is great! Sorry for the noise, but I am demonstrating commenting from the command line.

[cphyc]

Quick update: we can now do:

import yt
ds = yt.load("output_00080")

# Quickly define new fields
ds.fields.gas.double_density = ds.fields.gas.density * 2

# Quickly filter on the grid
ad = ds.all_data()
left = ad[ds.fields.gas.x < (0.5, "unitary")]

# Quickly filter particles
left_particles = ds.fields.io.particle_position_x < ds.quan(0.5, "unitary")
ds.add_particle_filter(left_particles)
ad[left_particles.name, "particle_position_x"]

[cphyc]

Note to reviewers: I'm waiting to get approval (if any :D) before writing the documentation. If you'd like to see more examples, I can write the doc before getting approval though.

[jzuhone]

Could you leave a comment as to why this is commented out?

[cphyc]

Ah good point! I went ahead and implemented this feature.

The annoying point is that defining an __eq__ method forces you to also implement a __hash__ method (which I did).
But this is worth it, since now we can write something like

ad = ds.all_data()
ds.fields.index.is_level_0 = ds.fields.index.grid_level == 0
ad["index", "is_level_0"]

low_res_only = ad.cut_region(ds.fields.index.grid_level == 0)

[chrishavlin]

For what it's worth at this point, I like the new method! I did check out the branch and quickly played around with defining fields and it all worked as expected but I haven't had time to actually review the code (and probably won't in the short term...). So +1 conceptual approval from me, and don't hold this up on my behalf if anyone else has time to get to it.

P.S., The most recent test failure does look related to the most recent changes.

[cphyc]

I found the source of the failures @chrishavlin was talking about. Here's the log of my little adventure:

• Defining the __eq__ forces us also to define a __hash__ function for DerivedFields to be hashable (e.g., to make sets of them, or use them as an entry to dictionaries),
• Since we support the syntax ds.fields.ftype1.fname1 == ds.fields.ftype2.fname2, which returns a DerivedFieldCombination, we need to evaluate the truthness of this expression, i.e. implement a __bool__ method on DerivedFieldCombination,
• That boolean should be True in all cases (since an object in Python is considered as True), unless we're comparing a field to another i.e.

# The following two evaluate to the same boolean
bool(ds.fields.ftype1.fname1 == ds.fields.ftype2.fname2)
ds.fields.ftype1.fname1 is ds.fields.ftype2.fname2

[cphyc]

@matthewturk @jzuhone @chrishavlin this should be ready to go now, tests (if they pass) + documentation included.