Refactor pipelines into classes

[talmo]

This PR constitutes a massive refactor started in #45 and #50 to fix how we compute our traits in a pipeline form.

While there are many changes, the key differences are that we now have the trait computation graph fully defined and structured as a set of classes in sleap_roots.trait_pipelines. They are organized into:

• TraitDef: This defines the concept of a trait, including which other traits are used to compute it, its name, the function that is used to compute it, and additional metadata such as whether it's scalar (i.e., needs to be summarized) and whether it should be included in output CSVs. This allows us to be flexible in defining intermediate traits that may not be included in final analyses but that are necessary to compute other traits. Fully defining the inputs and outputs also allows us to compute all of the traits in the appropriate order by enforcing a topological ordering of the computation graph.
• Pipeline: This is a base class which implements the generic trait computation steps, including frame-level, plant-level, and batch-level steps. Subclasses of Pipeline (such as DicotPipeline) can be defined by just inheriting from this class and defining two functions:
  • define_traits: Returns a list of TraitDef that defines the set of traits computed by the pipeline.
  • get_initial_frame_traits: Returns a dictionary of initial traits derived from the raw keypoint data. This is necessary because different pipeline types will have different initial traits depending on which root types are tracked (e.g., primary + lateral, primary + main, only main, etc.).

An example of how these are used:

from sleap_roots import Series, DicotPipeline
plant = Series.load(r"tests\data\soy_6do\6PR6AA22JK.h5", primary_name="primary_multi_day", lateral_name="lateral__nodes")
pipeline = DicotPipeline()

# Plant-level traits
traits = pipeline.compute_plant_traits(plant)
assert traits.shape == (72, 115)

# Batch level traits
all_traits = pipeline.compute_batch_traits([plant])
assert all_traits.shape == (1, 1018)

[codecov]

Codecov Report

Merging #51 (e71e8d9) into main (1d30dbd) will decrease coverage by 4.51%.
The diff coverage is 86.15%.

@@            Coverage Diff             @@
##             main      #51      +/-   ##
==========================================
- Coverage   81.40%   76.90%   -4.51%     
==========================================
  Files          13       13              
  Lines         726      762      +36     
==========================================
- Hits          591      586       -5     
- Misses        135      176      +41     

Files Changed	Coverage Δ
sleap_roots/convhull.py	76.11% <73.33%> (-6.74%)	⬇️
sleap_roots/points.py	78.57% <75.00%> (-18.49%)	⬇️
sleap_roots/lengths.py	76.92% <76.92%> (ø)
sleap_roots/networklength.py	81.48% <79.66%> (-11.06%)	⬇️
sleap_roots/tips.py	85.18% <81.81%> (-8.57%)	⬇️
sleap_roots/bases.py	88.46% <89.15%> (-3.13%)	⬇️
sleap_roots/angle.py	91.66% <89.74%> (-8.34%)	⬇️
sleap_roots/scanline.py	88.23% <92.30%> (-6.51%)	⬇️
sleap_roots/trait_pipelines.py	93.33% <93.33%> (ø)
sleap_roots/__init__.py	100.00% <100.00%> (ø)
... and 2 more

... and 1 file with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[linwang9926]

Delete fraction as not used in this function

[linwang9926]

Delete the fraction