int-brain-lab / ibleatools / 39670046212
51%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 52
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 56.103% (+1.4%) from 54.676%
39670046212

push

github

web-flow 
Interp/2b unit-level encoder (#146)

* unit_level_encoder: hierarchical GMM unit-level model, served by load_pretrained

The unit-level interpolation model as released for the arXiv paper (Hub tag 2026_W42):
- Waveform, 3D ACG and population-coupling autoencoders feed a 60-D latent (UnitAutoencoder).
- A hierarchical Gaussian mixture over the standardised latents: 30 full-covariance types,
  each split into up to 10 sub-types, with top-down assignment P(k,j|z) = P(k|z) P(j|k,z)
  (hierarchy.HierarchicalMixture).
- Mixture weights conditioned on the molecular context by two networks, one for the types and
  one for the sub-type shares within each type (context_weights.HierarchicalWeightModel).
- Members readout: each sub-type's phenotype is the responsibility-weighted mean of its TRAIN
  units' waveform features (hierarchy.MemberReadout).
- Data preparation tied to the channel release's context volumes and probe split; units are
  described by the channel list's waveform features defined for a mean waveform.
- UnitEncoder (models.unit_encoder) serves a release through load_pretrained: predict,
  sample, mixture_weights, encode, posterior, assign, selftest.
- The previous MultimodalAutoencoder + PointTransformerGMM wrapper is removed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* CHANGELOG: the unit-level interpolation model

---------

Co-authored-by: asaguy <as7709@columbia.edu>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

4284 of 7636 relevant lines covered (56.1%)

0.56 hits per line

Coverage Regressions

Lines Coverage ∆ File
63
35.66
4.5% unit_level_encoder/data.py
31
87.25
2.97% models/unit_encoder.py
24
66.67
20.72% unit_level_encoder/model.py
2
98.26
0.11% model_registry.py
1
99.22
-0.78% unit_level_encoder/config.py
Jobs
ID Job ID Ran Files Coverage
1 39670046212.1 52
56.1
GitHub Action Run
Source Files on build 39670046212