ECG Dataset - #53
Open
atharvas wants to merge 76 commits into
Open
Conversation
…se for the same. Plus minor edits to remove extra print statements.
Warning: This version does not support backpropogation to probabilities. Need to use gumbel softmax for that.
atharvas
marked this pull request as draft
March 15, 2024 00:02
atharvas
marked this pull request as ready for review
March 20, 2026 01:04
compute_metrics short-circuited to {neg_l1_dist, neg_l2_dist} for any
float-dtype labels, making classification metrics (F1, hamming, AUROC)
structurally unreachable for multi-label tasks (whose labels must be
float for BCE loss). Distinguish binary-indicator multi-hot matrices
from continuous regression targets so multi-label routes correctly;
regression path is unchanged. Add macro_auroc as a first-class
validation metric with per-column skip on degenerate single-class
columns, and register it in the NEARTrainerConfig allowlist.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
phase1_typed_ecg_dsl gains 5 new kwargs (disable_add, disable_mul,
disable_ite, disable_linear, disable_embeds) so each combinator and
per-group embed can be omitted independently for ablation. Each gates
the corresponding dslf.production call; defaults preserve T1 behaviour
exactly.
benchmark_reduced_attention_ecg.py exposes matching CLI flags
(--disable-{add,mul,ite,linear}, --disable-embeds a,b,...) plus
--train-seed to sweep data-shuffling seeds.
Add REPORT.md with (a) T1 headline reproduction (bit-exact 0.9017),
(b) 7-arm production ablation (mul is the single load-bearing operator;
add/ite are removable under budget), and (c) a reviewer-driven sweep
covering seed sensitivity, feature-group importance (amp features alone
= 0.899), depth/cost/hidden/LR sweeps, and neural-relaxation
minimisation (20 search epochs or hidden=16 both match T1). 26
experiments, ~80 GPU-hours.
Also extend CASE_STUDY.md (§5.6 multi-label evaluation) and
EXPERIMENTS.md (Phase 5) with the framework-defect narrative and M1-M4
multi-label results.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This adds the ECG dataset to the codebase along with a tutorial. This is a pretty messy update and still needs some cleanup (hence the draft).
Remaining TODOs: