AMR: Expand particle deposition onto the grid to vector fields. #5227
Conversation
|
Just to add, I am currently only considering |
chrishavlin
added
enhancement
|
@cphyc some octree enhancements/changes here that you might be interested in checking out when you're back from holiday :) |
|
@AnatoleStorck would be happy to look more at this after the test failures are addressed. |
|
I'd love to see this functionality in yt. Let me know if there are ways I can assist in testing this or debugging. |
| return self.oct_handler.domain_ind(self.selector) | ||
|
|
||
| def deposit(self, positions, fields=None, method=None, kernel_name="cubic"): | ||
| def deposit(self, positions, fields=None, method=None, kernel_name="cubic", vector_field=False): |
There was a problem hiding this comment.
You're missing the corresponding item in the docstring to explain what vector_field means.
yt/geometry/oct_container.pyx
Outdated
|
|
||
| source = source.reshape(np.shape(source) + (dims,)) | ||
| # If source is a vector field, we need to handle it differently. | ||
| is_vector_field = len(np.shape(source)) > 5 |
There was a problem hiding this comment.
Maybe use source.ndim == 5 instead?
yt/geometry/oct_container.pyx
Outdated
| source = source.reshape((source.shape[0], source.shape[1], | ||
| source.shape[2], source.shape[3], dims)) | ||
| if is_vector_field: | ||
| dest = np.zeros((len(dest), np.shape(source)[-2], dims), |
There was a problem hiding this comment.
I am not sure this is correct - if dest is not None, we should modify it in place instead of creating a new array. It is possible that the upstream code relies on dest being modified (if passed).
In addition, np.shape(source)[-2] is cryptic and requires knowing about https://github.com/yt-project/yt/pull/5227/files#diff-3cef3e330a489f3d2975d39b0f17b55a03f3134b27492604af30bc23f1bf7fa7R179. What about something like:
*_, ncells, ndim = source.shape
# assert ndim == dims?
dest = np.zeros((len(dest), ncells, dims), dtype=source.dtype, order='C')
yt/geometry/oct_visitors.pyx
Outdated
| #print("IN COPY2DARRAYF64") | ||
| #print(self.dest.shape, self.source.shape) |
There was a problem hiding this comment.
| #print("IN COPY2DARRAYF64") | |
| #print(self.dest.shape, self.source.shape) |
| # even though in other instances it varies the fastest. To see this in | ||
| # action, try looking at the results of an n_ref=256 particle CIC plot, | ||
| # which shows it the most clearly. | ||
| return ((i*dims[1])+j)*dims[2]+k |
There was a problem hiding this comment.
Shouldn't this be (i*dims[0] + j) * dims[1] + k?
| arr2.shape = arr2.shape + (1,) * (naxes - arr2.ndim) | ||
| return arr2 | ||
|
|
||
| cdef class ParticleDepositOperation: |
There was a problem hiding this comment.
Could it superclass the one defined in particle_deposit.pyx to minimize code duplication?
|
Note for @AnatoleStorck and @cphyc, the |
This works, but is still using the 'particle_deposit_vector.pyx' code. The latter is a duplication from 'particle_deposit.pyx', which only works for scalar fields but could be replaced with the more generic vector version.
Assume that we're depositing vector, and treat scalar fields as a special case of vectors
cphyc
force-pushed
the
add_vector_field_support_to_deposit
branch
from
34fba04 to
3526408
Compare
|
@chummels this should support SPH datasets, but I have never used an SPH dataset, and I'm not sure what the particle deposition would mean for SPH? Onto what are you deposing? Given my ignorance, I haven't really tested this, but it'd be amazing if you could give it a go |
|
We still have failures for AMR (non-octree) datasets. It seems that vector fields end up being stored vector-first, i.e., with shape |
5 participants
PR Summary
Currently, only scalar quantites of particles can be deposited to the grid (such as mass), with final shape (Ncells,) I want to be able to deposit a vector quantity (stellar spectra of each star particle) onto the grid with shape (Ncells, spectra_length), with the main goal to build spectra for each cell (nebular continuum + emission lines, and stellar spectra.) I have not been able to make the deposition of the vector fields work yet, so any help is greatly appreciated.
PR Checklist