Model time features drop sub-second precision
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 3/5
- Estimated time
- Half a day
- Newbie friendliness
- 55/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- python
- Domain
- machine-learning
Research direction
The floor division by NANOSECONDS_IN_SECONDS is at packages/kumo-core/src/kumo_core/encoders/tensor_frame_encoder.py lines 213 and 223, and packages/kumo-core/src/kumo_core/encoders/tensor_mapper.py line 777 does the same with 10**9. Start by checking how timestamp_unit reaches these encoders, then keep the sub-second remainder. Done means two events 300 ms apart get different time features on a millisecond index, and seconds-index encoder outputs stay bit-identical.
Written by the indexing model from the issue text.
Description
Problem
On an index with a sub-second clock (graph.snapshot(timestamp_unit="milliseconds"), #123), two events 300 ms apart are visible, ordered and windowed correctly, but the model still sees them at the same moment.
Cause
The encoders turn every timestamp into whole seconds before building features:
packages/kumo-core/src/kumo_core/encoders/tensor_frame_encoder.py:213seed_time // NANOSECONDS_IN_SECONDSpackages/kumo-core/src/kumo_core/encoders/tensor_frame_encoder.py:223time.view(-1, 1) // NANOSECONDS_IN_SECONDSpackages/kumo-core/src/kumo_core/encoders/tensor_mapper.py:777unix // 10**9
So relative-time and timestamp features are identical for events within the same second.
Proposed fix
Keep the sub-second part when computing time features, for example by dividing as floats or carrying the remainder. On a seconds index every timestamp is a whole second, so the model's inputs stay exactly as today.
Acceptance
- On a millisecond index, two events 300 ms apart get different time features.
- On a seconds index, encoder outputs are bit-identical to today's.
- Dominant language
- Python
- Stars
- 20
- Forks
- 6
- Avg merge
- 1d 21h
- Merged PRs (30d)
- 41
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from NVIDIA/kumo-engine
-
enhancement
Difficulty 4/5 3-5 days Newbie friendliness 42/100
NVIDIA/kumo-engine#74 ·
Maintainers usually reply within 1 day
All issues in NVIDIA/kumo-engine
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 66/100
microsoft/agent-framework#9946 ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 Under an hour Newbie friendliness 88/100
btclib-org/btclib-node#2198 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
Maintainers usually reply within 3 days
-
Remove the em dashes from the skill smart-account referencePossibly taken @Vamsi-2581 claimed this today. Openagent-ready area: docs documentation good first issue
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
Soneso/stellar-agent-wallet#541 · 2 comments ·
Maintainers usually reply within 1 day
-
good first issue
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
Maintainers usually reply within 1 day