Project 5: Does training cost you sleep?

Goal: a script of your own that puts two lists side by side, your workouts and your nights, and answers a question your apps answer separately: do you sleep less after a hard day?

Time: 30 minutes. Level: some Python. Needs: project 2, with workouts from a watch, Strava, Polar, Suunto, intervals.icu or the Health app, and sleep from any source. They do not need to come from the same device.

1. Start from the connection blockCopy Anchor Link

awk '/^def nights_by_app/{exit} {print}' recipes/sleep-report.py > training-vs-sleep.py

2. Join workouts and nightsCopy Anchor Link

Paste this at the end of training-vs-sleep.py:

from datetime import datetime, timedelta, timezone
from zoneinfo import ZoneInfo


def local(stamp, record):
    """A UTC time from the API, on the clock of the place it was recorded."""
    t = datetime.fromisoformat(stamp[:19]).replace(tzinfo=timezone.utc)
    zone = record["metadata"].get("timezone")
    return t.astimezone(ZoneInfo(zone)) if zone else t.astimezone()


since = (datetime.now() - timedelta(days=56)).date()

# Minutes of training per day.
training = {}
for w in records("workouts", start=since.isoformat()):
    day = local(w["start_date"], w).date()
    training[day] = training.get(day, 0) + (w["duration_s"] or 0) / 60

# Hours of sleep per night, named by the morning it ends on.
sleep = {}
for s in records("sleep", start=since.isoformat()):
    if s["bedtime_end"]:
        morning = (local(s["bedtime_end"], s) + timedelta(hours=10)).date()
        sleep[morning] = sleep.get(morning, 0) + (s["total_sleep_seconds"] or 0) / 3600

# The night after each day: trained 45 minutes or more, or rested.
groups = {"after 45+ min of training": [], "after a rest day": []}
for morning, hours in sleep.items():
    minutes = training.get(morning - timedelta(days=1), 0)
    if minutes >= 45:
        groups["after 45+ min of training"].append(hours)
    elif minutes == 0:
        groups["after a rest day"].append(hours)

print(f"Sleep since {since:%d %b}")
for name, nights in groups.items():
    if nights:
        print(f"  {name:27} {sum(nights) / len(nights):.2f} h   ({len(nights)} nights)")

python3 training-vs-sleep.py

Sleep since 18 Aug
  after 45+ min of training   6.76 h   (18 nights)
  after a rest day            7.16 h   (30 nights)

How it works.

1. records("workouts") and records("sleep") read the two lists. A workout from Strava and a night from an Oura ring have the same shape here, so the script never asks where they came from.
2. Both are put on your local clock. The API sends UTC; metadata.timezone says where a record was made when the source knows it, otherwise the script uses your computer's zone.
3. Each night is matched with the day before it: the night that ends on Tuesday morning follows Monday's training.
4. Days with a short workout (under 45 minutes) are left out of both groups, so the comparison is between clearly hard days and clearly easy ones.

Read it with care. A difference over a handful of nights can be chance, and a hard day can fall on the day you also went out. The count next to each average is there to remind you: the more nights, the more it means.

Make it yoursCopy Anchor Link

• Late sessions. Split the training days by the hour the workout started: before or after 18:00. Evening sessions are often the ones that shorten a night.
• HRV, not hours. Use records("daily", start=...) and its hrv_avg_ms (rings and bands) in place of sleep hours, matched the same way.
• By sport. Each workout has an activity_type (running, cycling, swimming...): keep one and see if the answer changes.

Share itCopy Anchor Link

Your script and what you found, even "no difference at all", in Show and tell. A finding that holds over months is worth a post of its own.

Next: Project 6, Your own dashboard.