int-brain-lab / ibleatools / 39609553608
51%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 46
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: 54.676% (-7.5%) from 62.159%
39609553608

push

github

web-flow 
Interp/2a spatial encoder (channel-level model) (#145)

* spatial_encoder: 2026_W39 feature list, denoised MERFISH context, release statistics

The channel-level interpolation model as released for the arXiv paper:
- FEATURE_LIST is the 43 features of the denoised table from vintage 2026_W39, grouped
  LF | AP | waveform (FEATURE_GROUPS), with the first spatial difference under its
  bipolar-referencing name (*_bpr, #138).
- The MERFISH context PCA is fitted on the denoised subclass cell-type volume
  (iblatlas.genomics.merfish.denoise_volume; iblatlas>=1.2.0).
- SpatialEncoder exposes .features, .split(), .load_confidence_model(), .neighbor_bank() and
  the training statistics (preprocessing/channel_stats.npz), and mirrors right-hemisphere
  queries to the left hemisphere as in training.
- Position inputs are optional and off by default.
- The probe-confidence model trains in minutes: one shared Allen atlas, cached synthetic samples.
- examples/inference_spatial_encoder.py moves to paper-ephys-atlas.

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

* SpatialEncoder: releases listing *_csd_diff1 load under the *_bpr names

Releases trained before the bipolar-referencing rename (2026_W41 and earlier) list the first
spatial difference as *_csd_diff1, while the feature tables are now read with *_bpr
(data.LEGACY_FEATURE_RENAMES). The wrapper reports their features, prediction columns and
golden example under the current names; the manifest's list and digest are unchanged.

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

* CHANGELOG: the channel-level interpolation model

* CHANGELOG: list only the example removed by this PR

* Address review: drop legacy feature names, load the denoised MERFISH volume, trim comments

---------

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

3461 of 6330 relevant lines covered (54.68%)

0.55 hits per line

Coverage Regressions

Lines Coverage ∆ File
28
84.08
84.08% models/encoder_inpainting.py
2
98.15
0.02% model_registry.py
Jobs
ID Job ID Ran Files Coverage
1 39609553608.1 46
54.68
GitHub Action Run
Source Files on build 39609553608