Repository navigation
Conversation
Previous review (2026-11-09)Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting. Reviewed at head This adds a useful missing-data warning, and interior gaps are reported correctly, but gaps at the edges of the analysis window are never counted. Should fix
Short batches that the FFT skips are counted as covered, but that is existing FFT behaviour, not missing data, so it is not raised as an issue. |
Clip to the time covered by all IMUs rather than this IMU's own data, so an IMU that starts late, stops early or has no data in the window is counted as missing.
Previous review (2026-11-09)Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting. Reviewed at head This PR now looks good to merge: the missing-data warning is correct for interior gaps and for gaps at the edges of the analysis window. Since the last review, 34bde32 clips to the shared Gyro_batch.start_time/end_time, which fixes the edge-of-window false negative (FilterReview/FilterReview.js:299). An IMU that starts late, stops early or has no data in the window now reports 50%/100% instead of 0%, and continuous data still shows 0%. The earlier note about short batches the FFT skips still stands as a non-issue. Not fully reviewed: tested only with synthetic gyro data, not with a real raw-IMU .bin log loaded in a browser. |
IamPete1
left a comment
There was a problem hiding this comment.
I like the idea, but I think this needs to recalculated on the FFT object because it can change with window size. Smaller batches are less likely to run into missing bits. Larger batches can waste a lot of data if there is not quite enough room for another whole window.
So the end result would be a percentage of usable data which would be some percentage from the "batching" and then a calculation on the FFT coverage of the batches based on the window size.
I think we could always show the number, but only make it red if there is lots of dropped data. This also means the gyro boxes all stay the same height.
The share of the analysis time used is now taken from the FFT windows rather than the raw data sections, so it includes the data lost at the end of each section that is too short for another whole window, and changes with the window size. It is always shown for raw logs and turns red below 90%, the gyro boxes keep the same height.
|
Reworked: it's now "Usable data", the share of the analysis time covered by the FFT windows used, so it includes what's lost at the end of each section and changes with window size. Always shown for raw logs, red below 90%, so the boxes keep the same height. |
Previous review (2026-11-20)Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting. Reviewed at head This rework looks right, and it is worth merging once one small performance fix is in. Since the last review, f6d04e8 replaced the missing-data figure with "Usable data" from the FFT windows, as IamPete1 asked. The figure now follows window size and includes the end-of-section loss, and the earlier edge-of-window fix still holds. Should fix
Minor (2)
The earlier short-batch note no longer applies: short sections now count as unusable by design. Not fully reviewed: no real raw-IMU .bin log was loaded through the parser (all testing used synthetic gyro data), and the eslint CI check was still pending for this head. |
Previous review (2026-11-20)Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting. Reviewed at head This looks good to merge: the usable-data figure is correct and nothing is left open. Since the last review, cd2dde5 fixed all three earlier points. The window range is now found once before the loop (FilterReview.js:321), removing the quadratic scan. The red colour now follows the rounded percent that is shown (FilterReview.js:296). The stray Not fully reviewed: no real raw-IMU .bin log was loaded through the parser or in a browser at this head (all testing used synthetic gyro data), and the eslint CI check is still pending (a local ESLint run passes). |
| Logging rate: <label id="Gyro0_FFT_infoA"></label> Hz<br><br> | ||
| Frequency resolution: <label id="Gyro0_FFT_infoB"></label> Hz | ||
| Frequency resolution: <label id="Gyro0_FFT_infoB"></label> Hz<br><br> | ||
| <label id="Gyro0_usable" style="visibility:hidden">Usable data:</label> |
There was a problem hiding this comment.
done, it's display none now until there's a value
Hidden rather than invisible, so batch logs don't get a blank line.
|
Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting. Reviewed at head This looks good to merge, and nothing is left open. Since the last review, f32f794 hides the Usable data line until it has a value, as asked. The wrapper span starts as display:none (index.html:234) and is shown only when there is a figure (FilterReview.js:293), so the blank space is gone and batch logs look as they did before. Not fully reviewed: no real raw-IMU .bin log was loaded through the parser in a browser (testing used synthetic gyro data), and the CI result is taken from job metadata (a local ESLint run passes). |

"Warn if lots of missing data" from #10.
For raw IMU logs, each gyro's info now shows "Usable data: N% of analysis time", the share of the analysis time covered by the FFT windows that are used. Data is lost where it is missing from the log (the raw loader splits the data into sections at gaps), and at the end of each section where there isn't room for another whole window, so the number changes with the window size. It is clipped to the time covered by all IMUs, so an IMU that starts late or stops early counts as lost.
The line is always there so the gyro boxes stay the same height, and turns red below 90% (
usable_data_warn_fraction, happy to change it).Batch logs are not shown. The batch sampler logs short bursts by design and waits rather than dropping data when the log buffer is full (
BatchSampler.cpp, "maybe later"), so the coverage there isn't lost data.Checked with continuous synthetic raw data over a 47 s analysis time at 1 kHz, with the expected window coverage: