Repository navigation
Conversation
Test CoverageTotal: 74.71% (21031 / 28151 lines)
|
Logiquo
force-pushed
the
feat/seq-v2-2
branch
from
October 30, 2026 06:04
7e4e75a to
8b7487c
Compare
Collaborator
Author
|
/review |
AI Reviewsuggestion: not available This PR is too large for AI review (over 300 files or 200000 diff characters), so it is labeled |
Logiquo
force-pushed
the
feat/seq-v2-2
branch
2 times, most recently
from
October 30, 2026 11:36
6912de3 to
b8522f4
Compare
Logiquo
changed the title
Logiquo
force-pushed
the
feat/seq-v2-2
branch
from
October 30, 2026 12:08
b8522f4 to
c347cb6
Compare
- TimeseriesLabelProcessor and its binary and regression subclasses are now restricted SequenceProcessorV2 fields: one series level with scalar values and strictly increasing times. A batch is packed as (values [ΣN, 1], time [ΣN], batch offsets [B + 1]); there is no grid or mask. - The timeseries losses average over the label records, and Trainer.inference splits the packed labels into one array per sample. _resolve_mode picks the most specific registered processor class, so the label subclasses resolve to their own modes. - RNN and MultimodalRNN reject per-step (timeseries_*) labels. - The five YAIB MIMIC-IV tasks use a sequence_v2 series for the dynamic features (forward fill, then mean, then zero imputation, z-score normalization and missing indicators) and per-hour label records. - Tests use a minimal per-step model (tests/core/label_record_model_utils.py) for the loss, Trainer and metrics.
Logiquo
force-pushed
the
feat/seq-v2-2
branch
from
October 30, 2026 12:40
c347cb6 to
ce890cd
Compare
This branch has not been deployed
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
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.
Second PR in the stack toward a unified sequence representation, on top of #1343. It moves the per-step (timeseries) labels and the YAIB MIMIC-IV tasks to
SequenceProcessorV2, and updates the loss andTrainerto match. No model consumes per-step V2 labels yet;Unified_RNNfollows in a later PR of this stack.Changes
TimeseriesLabelProcessor,TimeseriesBinaryLabelProcessorandTimeseriesRegressionLabelProcessorare now restrictedSequenceProcessorV2fields: oneserieslevel with scalar values and strictly increasing times. A batch is packed as(values [ΣN, 1], time [ΣN], batch offsets [B + 1]); there is no grid or mask.timeseries_*losses average over the label records, andTrainer.inferencesplits the packed labels into one array per sample for the timeseries metrics.BaseModel._resolve_modepicks the most specific registered processor class, so the label subclasses resolve to their own modes rather thansequence_v2.RNNandMultimodalRNNreject per-step labels.sequence_v2series for the dynamic features: forward fill, then mean, then zero imputation, z-score normalization and missing indicators. A test checks this matches the previousTimeseriesProcessorpreprocessing.Tests
tests/core/label_record_model_utils.pyadds a minimal per-step test model (one linear layer on each label's time), so the label processors, loss,Trainerand metrics are tested without depending on a particular architecture.