Skip to content

benchmark: q1 ground truth - #98

Closed
GnaneshGnani wants to merge 23 commits into
mainfrom
debs_q1
Closed

GnaneshGnani wants to merge 23 commits into
mainfrom
debs_q1

Conversation

@GnaneshGnani

@GnaneshGnani GnaneshGnani commented Apr 1, 2026

Copy link
Copy Markdown
Contributor

Q1 — EMA indicators (DDSketch p50 vs EMA38)

Adds ground truth computation and 60-minute accuracy benchmark for Q1: exponential moving average (EMA38) indicator tracking.

Benchmark results

Setup: Day 08-11-21, data_filtered, accuracy_sla=0.01, 12 × 5-min windows
Sketch: DDSketch, relative_accuracy=0.01, quantiles [0.0, 0.25, 0.5, 0.75, 0.9, 0.99, 1.0]
Metric: fraction of (symbol, window) pairs where |ema38 − sketch_p50| / |ema38| < 1%; threshold ≥ 95%.

Windows Pass rate Verdict
12 (09:00–10:00 CET) 11 / 12 ⚠️ One early market-open window fails

The 09:10–09:15 window (frac=0.943) is the only failure — EMA38 is still settling on initial large market-open price moves. All subsequent windows pass comfortably (0.952–0.967).

See results/benchmarking.md for per-window breakdown.

Comment thread datasets_eval/debs/benchmark/ground_truth/q1.py
Base automatically changed from debs_benchmark to main April 4, 2026 22:27
@zzylol

zzylol commented Apr 6, 2026

Copy link
Copy Markdown
Contributor

Please solve the merge conflict to main branch as well.

@GnaneshGnani
GnaneshGnani requested a review from zzylol April 6, 2026 23:54
@zzylol

zzylol commented Apr 13, 2026

Copy link
Copy Markdown
Contributor

Closing this PR because EMA should not be computed by sketches.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants