Extend TensorData with index_view() and mut_index_view()
#3880
Conversation
- Implement `ravel` method for computing row-major index offsets. - Add canonical index handling via `canonicalize_index`. - Include unit tests for validation.
- Implement `index_view` and `mut_index_view` methods for accessing/modifying tensor elements via indices. - Introduce `TensorDataIndexView` and `TensorDataIndexViewMut` structs for indexing support. - Add unit tests to validate index view functionality.
|
Maybe this should be |
Codecov Report❌ Patch coverage is
❌ Your project check has failed because the head coverage (65.02%) is below the target coverage (80.00%). You can increase the head coverage or adjust the target coverage. Additional details and impacted files@@ Coverage Diff @@
## main #3880 +/- ##
==========================================
+ Coverage 65.00% 65.02% +0.01%
==========================================
Files 1180 1180
Lines 139535 139622 +87
==========================================
+ Hits 90707 90784 +77
- Misses 48828 48838 +10 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
the view should maybe cache the slice, save some lookup time there for repeated calls? |
…ut_index_view`. - Update method names for improved clarity and consistency. - Adapt associated tests to reflect the changes.
…nates. - Update function signatures to replace array with slice usage. - Adjust related assertions and iteration logic accordingly. - Update associated unit tests to match the new API.
…exView implementations - Update `Index` and `IndexMut` trait implementations to use slice-based indices. - Adjust relevant indexing logic and tests to align with the new API.
There was a problem hiding this comment.
As we discussed on discord the other week, I am not really in favor of such extensions to TensorData.
Adding indexing makes it try to be both a container and a computational view, which is not its responsibility.
Such manipulations should be performed on Tensor. I understand that currently requires moving between backends since some like CUDA don't have a CPU device, which is a bit awkward and not so ergonomic. But we are going to provide a better multi-backend support that will provide proper CPU-backed Tensors that can be used for this sort of manipulation.
For now, I much prefer to add lightweight utilities to Shape (like the linked PR).
2 participants
This is built on #3879
Consider also
at()[[a, b, c]]andat_mut()[[a, b, c]]