samsung-health-sdk 0.2.4

Parse and analyse Samsung Health export data as pandas DataFrames

samsung-health-sdk

A Python SDK for parsing and analysing Samsung Health export data.

Load any health metric from a Samsung Health export directory as a pandas DataFrame with a single function call. Compare data across multiple people or time windows. Derive higher-level health features from the raw data.

Installation

pip install samsung-health-sdk

Quick Start

from samsung_health_sdk import SamsungHealthParser, SamsungHealthComparator

# Point at your Samsung Health export directory
p = SamsungHealthParser("path/to/samsunghealth_export_dir")

# See all available metrics
print(p.list_metrics())

# Load heart rate (hourly summaries or minute-level)
hr     = p.get_heart_rate("2024-10-01", "2024-10-31")
hr_min = p.get_heart_rate("2024-10-01", "2024-10-31", granularity="minute")

# All supported metrics
sleep  = p.get_sleep("2024-10-01", "2024-10-31")
skin   = p.get_skin_temperature("2024-10-01", "2024-10-31", granularity="minute")
stress = p.get_stress("2024-10-01", "2024-10-31")
spo2   = p.get_spo2("2024-10-01", "2024-10-31")
steps  = p.get_steps("2024-10-01", "2024-10-31")
hrv    = p.get_hrv("2024-10-01", "2024-10-31")
rr     = p.get_respiratory_rate("2024-10-01", "2024-10-31", granularity="minute")
ex     = p.get_exercise("2024-10-01", "2024-10-31")
mv     = p.get_movement("2024-10-01", "2024-10-31")  # per-minute activity_level

# Generic accessor for any metric by its full name
df = p.get_metric("com.samsung.shealth.vitality_score", start="2024-10-01")

Feature Engineering

HealthFeatureEngine derives meaningful higher-level features from the raw data:

from samsung_health_sdk.features import HealthFeatureEngine

eng = HealthFeatureEngine(p, tz_offset_hours=5.5)  # tz_offset_hours: your UTC offset

# Per-night sleep quality: efficiency, deep/REM %, fragmentation, composite score
sleep_stats = eng.sleep_sessions("2025-01-01", "2025-03-31")

# Per-night HRV + respiratory rate + movement restlessness during sleep
physio = eng.nightly_physiology("2025-01-01", "2025-03-31")
# Columns: rmssd_mean, rmssd_min, rmssd_std, rr_mean, rr_std,
#          restlessness_score, restless_min, hrv_suppression_flag

# HRV readiness: today vs your rolling N-day personal baseline
readiness = eng.hrv_readiness("2025-01-01", "2025-03-31", baseline_days=14)
# Columns: rmssd_mean, baseline_14d, deviation_pct, readiness_score, low_readiness_flag

# Previous-day stress deviation vs that night's sleep quality
impact = eng.stress_impact_on_sleep("2025-01-01", "2025-03-31")
# Uses stress deviation from rolling baseline, not absolute score

# Per-day activity breakdown + HR context + stress
profile = eng.daily_activity_profile("2025-01-01", "2025-03-31")
# Columns: sedentary_min, light_min, low_mod_min, moderate_min, vigorous_min,
#          active_min, mean_hr_active, median_hr_day, mean_stress, stress_deviation_pct

# Walking cardiac load trend (HR / speed — lower = more aerobically efficient)
cardiac = eng.walking_cardiac_load("2024-11-01", "2025-06-30", source="auto")
# source='auto': pedometer (most accurate) > movement (accelerometer-based,
#                extends to Nov 2024) > exercise summaries (Jun 2022+)
# Columns: date, duration_min, distance_m, speed_mps, mean_hr, cardiac_load,
#          source, rolling_4w_cardiac_load, cardiac_load_trend

Multi-Person Comparison

p1 = SamsungHealthParser("path/to/person1_export")
p2 = SamsungHealthParser("path/to/person2_export")

comp = SamsungHealthComparator({"Alice": p1, "Bob": p2})

# Compare heart rate — absolute calendar window
df = comp.compare_heart_rate("2024-10-01", "2024-10-31")

# Align to relative Day 0 per person (time_shift=True)
df = comp.compare_heart_rate("2024-10-01", "2024-10-31", time_shift=True)

Export Format

Samsung Health exports a directory containing:

• CSV files per health metric (some with a metadata row, some without — auto-detected)
• jsons/ subdirectory with per-minute binning JSON files
• files/ subdirectory with binary attachments (ECG waveforms, photos)

The SDK handles BOM encoding, namespaced column headers, UTC offset timestamps, trailing commas, and lazy JSON loading automatically.

Sleep Stage Codes

Code	Label
40001	Awake
40002	Light
40003	Deep
40004	REM

Movement Activity Levels

Per-minute accelerometer intensity from get_movement():

Range	Intensity
0–5	Sedentary
5–20	Light
20–50	Low-moderate
50–100	Moderate
100+	Vigorous

Requirements

• Python 3.9+
• pandas >= 2.0
• numpy >= 1.24

Contributing

git clone https://github.com/Devasy/samsung-health-sdk
cd samsung-health-sdk
pip install -e ".[dev]"
pytest