Skip to content

README.md

Latest commit

 

History

History
119 lines (91 loc) · 5.3 KB

README.md

File metadata and controls

119 lines (91 loc) · 5.3 KB
Raw

English | 中文

Sample Data

Two CSV files covering 14 consecutive days of ring history data, so you can build visualizations, analytics, and algorithms without wearing a device for two weeks first.

File Rows Interval Content
sample_14days_Measurement.csv ~7,900 2.5 min Heart rate, HRV, SpO2, respiration rate
sample_14days_Activity.csv ~1,300 15 min Battery, steps, active seconds, skin temperature

These files are synthetic, generated to match the statistical distributions and timing behaviour of real device captures. They are not recorded from an actual person. The column format is identical to what the app writes on History Data sync, so you can mix these files with your own exports and parse both with the same code.

Measurement — sample_14days_Measurement.csv

Header: time,hr,hrv,spo2,respRate,hrSuccess,spo2Success

Column Type Notes
time ISO8601 UTC e.g. 2026-07-01T00:02:30.000Z
hr int Heart rate (bpm). See "Reading the validity flags" below.
hrv int Heart rate variability. 0 when there is no valid reading.
spo2 int Blood oxygen (%). See "Reading the validity flags" below.
respRate int Respiration rate stored x10 — 124 means 12.4 breaths/min. 0 when invalid.
hrSuccess true/false Whether hr in this row is a real reading
spo2Success true/false Whether spo2 in this row is a real reading

Reading the validity flags — important

Rows are written every 2.5 minutes, but the sensors do not produce a value in every slot. This mirrors real device behaviour:

  • Heart rate is measured roughly every 5 minutes, so about half the rows carry no real HR.
  • SpO2 is measured roughly every 15 minutes. While the wearer is awake the cadence drifts (a heart-rate measurement can preempt it); during sleep it is steady.

When a slot has no reading, the value column contains the device's placeholder 18 and the matching *Success column is false.

Always filter on the flags rather than the values (the 18 placeholder is not a real measurement):

# Python / pandas
df = pd.read_csv('sample_14days_Measurement.csv')
# pandas parses the true/false column as a real boolean, so it can filter directly:
valid_hr = df[df.hrSuccess]['hr']            # real heart-rate readings
valid_spo2 = df[df.spo2Success]['spo2']      # real SpO2 readings
resp_bpm = df[df.spo2Success]['respRate'] / 10   # respRate is stored x10
// Dart — read the CSV and keep only rows with a real heart-rate reading.
// Each line is: time,hr,hrv,spo2,respRate,hrSuccess,spo2Success
final lines = await File('sample_14days_Measurement.csv').readAsLines();
final validHr = <int>[];
for (final line in lines.skip(1)) {          // skip the header row
  final c = line.split(',');
  if (c[5] == 'true') validHr.add(int.parse(c[1]));   // c[5]=hrSuccess, c[1]=hr
}

Activity — sample_14days_Activity.csv

Header: time,batteryPercent,steps,activeSeconds,temperaturesC

Column Type Notes
time ISO8601 UTC 15-minute interval
batteryPercent int 0–100 Ring battery
steps int Steps within this 15-minute interval (a delta, not a running total)
activeSeconds int Active seconds within this interval (0–900; 900 = the full window)
temperaturesC float list 15 skin-temperature samples (one per minute), semicolon-separated, e.g. 33.4;33.5;33.2 
# Daily step totals
df = pd.read_csv('sample_14days_Activity.csv', parse_dates=['time'])
daily = df.groupby(df.time.dt.date)['steps'].sum()   # ~7k-12k per day

# Skin temperature: expand the semicolon-separated samples
temps = df.temperaturesC.str.split(';').explode().astype(float)

What the data looks like

Realistic patterns are built in, so charts and algorithms produce sensible results:

  • Circadian heart rate — lower during sleep (~58 bpm), higher during the day, with occasional activity peaks. Valid range 51–107.
  • SpO2 — mostly 94–99, with mild dips during sleep. Valid range 86–99.
  • HRV — inversely tracks heart rate, higher during sleep. Range 21–104.
  • Steps — concentrated in daytime, near zero overnight, ~7k–12k per day.
  • Battery — drains over several days and recharges periodically.
  • Gaps — the ring is off the finger while charging (~1 hour, every other day), so there are no rows at all in those windows. Real syncs look like this too — make sure your charts handle missing time ranges.

All values stay within physiologically plausible ranges. Real captures occasionally contain sensor glitches (implausible temperatures, out-of-range heart rates); those are deliberately not reproduced here so they don't look like bugs in your own code.

Combining the two files: they share the same UTC time axis but different intervals (measurement every 2.5 min, activity every 15 min). To analyse them together — e.g. "was heart rate elevated during high-step periods?" — resample or bucket both to a common window (15 min works well) on the time column before joining.

Large files: prefer streaming/chunked reading if your tooling struggles with the measurement file (~380 KB, ~7.9k rows).