Plot your Fitbit data in Python (API v1.2)

Use the fitbit api v1.2 to access your health sleep heart data


Sleeping, and python. Two of my favourite things, when combined with the the Python Fitbit library, Matplotlib, and Pandas, can generate informative plots of your sleeping habits! This post explores how we can pull date from the Fitbit API, create a Pandas Dataframe, and then plot the results.

In this tutorial, I’ve used Python 3.7 to access data, the Fitbit Python library (with some adjustments for v1.2 of the Fitbit API), Pandas to manage the data, and Matplotlib to generate some visualisations.

Fitbit Direct Export Functionality

Note that you can also request data as a single dump directly from the Fitbit website. Log in, navigate to settings, data export, and “request export”. (see instruction here) Note it may take some hours to export your data if there is a lot there, I may add some scripts to parse data in this format if I can.

Two years of my Fitbit sleep data showing my total time in bed vs my actual time sleeping. Fitbit would suggest that I have a sleep “efficiency” of about 87.5%! Generate your version of this graph with the code below.

The Fitbit API and OAuth access

The Fitbit API is a well-documented RESTful API where all of your movement, sleep, heart rate, and food logging data (if tracked by a Fitbit device) can be programmatically accessed. To access data, you’ll need to create an “application” with a registered Fitbit developer user account. The applications will be granted access to your personal Fitbit data through the API. Application authorisation follows what’s called the “OAuth2.0” specification.

There are different API endpoints for each type of data, and all of the responses come back in JSON formats that can be easily converted back to Python dictionaries and Pandas Dataframes for processing.

Example JSON response:

  "logId" : 27198605014,
  "dateOfSleep" : "2020-05-16",
  "startTime" : "2020-05-16T03:01:00.000",
  "endTime" : "2020-05-16T11:10:30.000",
  "duration" : 29340000,
  "minutesToFallAsleep" : 0,
  "minutesAsleep" : 437,
  "minutesAwake" : 52,
  "minutesAfterWakeup" : 0,
  "timeInBed" : 489,
  "efficiency" : 97,
  "type" : "stages",
  "infoCode" : 0,
  "levels" : {
    "summary" : {
      "deep" : {
        "count" : 4,
        "minutes" : 73,
        "thirtyDayAvgMinutes" : 77
      "wake" : {
        "count" : 40,
        "minutes" : 52,
        "thirtyDayAvgMinutes" : 57
      "light" : {
        "count" : 41,
        "minutes" : 264,
        "thirtyDayAvgMinutes" : 254
      "rem" : {
        "count" : 10,
        "minutes" : 100,
        "thirtyDayAvgMinutes" : 103

Oauth2 primer

To make our lives easy, we’ll use the existing and excellent Fitbit Python library to handle some of the over and back for application authentication, but it’s worth understanding the basics. OAuth2 is a method of granting applications or users access to certain data or resources, and essentially the flow is:

  1. Your application points the end user (through a browser) to the Fitbit authorisation page using an application client ID and secret key.
  2. The user logs in and “grants access” to different “scopes” (descriptions of data access levels).
  3. An “authorisation code” is generated which is given back to your application by redirecting the browser to a “callback url” (for now, this will be hosted on your computer).
  4. The application creates an “access token” and “refresh token” based on the authorisation code.
  5. The “access token” is used in further API requests to get data.
  6. Over time, the access token will expire (8 hours), and the refresh token can be used to create a new access and refresh token.

See here for Fitbit specific implementations of OAuth2.

Fitbit Versa SmartWatch. The Fitbit data for this tutorial came from my own Fitbit Versa watch.

Create a personal Fitbit App

Start off by creating a new application on

Set up an account at and create a new “application”. You’ll be creating just a “personal” application for use only with your own account, so the details for website, terms of service, and privacy policy don’t particularly matter.

The key element to get right here is to set the “Callback URL” to “”. This URL will be used as part of the OAuth2 authorisation described above in step 3. My application setup looks like this:

Fitbit Personal Application Setup. These parameters are used to set up a Personal Fitbit app so that I can download my sleep data from the Fitbit API using Python. Note that the Callback URL needs to be set to

When the application is created, you’ll be able to download the Application “client ID” and “client secret”. These are the core credentials for your application, and are used during the OAuth2 process – store these somewhere for the next step.

Authenticate and get an API token

There’s a useful file in the Fitbit Python library called “” (original available here). This file implements a small local web server that you can use to authenticate with the Fitbit API to request an access and refresh token.

There’s a few simple steps to run this application locally on your machine:

  1. Copy the contents of to a local file with the same name.
  2. Install the cherrypy (pip install cherrypy) and Fitbit (pip install fitbit) libraries into your Python environment. (Ideally, set up a local virtual environment for your project) Cherrypy is a lightweight web server software that helps with the redirect URL from Fitbit during the OAuth2 flow.
  3. Activate your virtual environment, change to the directory where you have created your file and run the file, specifying your application client ID and secret key in the command, e.g:
    for me, this was similar to (not real codes here):
    python 22ARNQ f3ad3eedfabccbd33a3149f3b
  4. The script will open a browser window that presents a login page to Fitbit, where you can give permission to your application to access your data.
  5. When successfully authenticated, the script will print the access token and refresh token to the console window.
Login page presented when running the file. The script directs you to the Fitbit authentication page using your application client ID and secret key, and generates access tokens when you successfully log in.
Depending on the data that you set up with your personal Fitbit application, these permissions may vary. Choose the permissions for the data that you’d like to access.
The file outputs your access and refresh tokens once authentication is complete on the Fitbit page in your browser. Record these for use with your API data requests later.

Download your sleep data

The Fitbit python library, by default, is set up to use version 1 of the Fitbit API, whereas there is better quality data available from the V1.2 API. We’re going to use the Fitbit library to manage the data requests, but we’re going to manually specify the V1.2 API URLs so that we get the best data.

For this particular post, we’re interested in sleep data, and the resource URLs that are useful are:

  • Per day data available at:[user-id]/sleep/date/[date].json
  • Date range data availabile at (max 100 days):[user-id]/sleep/date/[startDate]/[endDate].json

The response from the API, when you are authenticated, includes summaries of your sleep for each date:

  • A breakdown of your night, in two formats:
    • “Stages”: This is more detailed analysis of your sleep on the latest devices. The data includes total event count and total minutes of deep, light, REM, and wakefulness for the night
    • “Classic”: A more basic breakdown of your sleep into just asleep, awake, and restless moments.
  • Summary stats on total minutes awake, total minutes asleep, time in bed, minutes to fall asleep (I found this last one slightly unreliable).

The data is split in two sections in the JSON response, “data” which includes any sleep stages or wake periods longer than three minutes, and a “shortData” which includes short wake periods less than 3 minutes in duration.

Here, I want to download more than 100 days of data, and so we need to generate all of the 100 day periods between our ultimate date targets. I’ve used the Arrow date/time management library to make this easier.

import fitbit
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
import arrow.  # Arrow is a really useful date time helper library

client = fitbit.Fitbit(
    '<paste application client ID>', 
    '<paste application secret key>',
    access_token='<paste access_token here>', 
    refresh_token='<paste refresh token here>'

start_date = arrow.get("2018-09-01")
end_date = arrow.get("2020-11-01")

# Create a series of 100-day date-range tuples between start_date and end_date
date_ranges = []
start_range = start_date
while start_range < end_date:
  if start_range.shift(days=100) < end_date:
    date_ranges.append((start_range, start_range.shift(days=100)))
    start_range = start_range.shift(days=101)
    date_ranges.append((start_range, end_date))
    start_range = end_date

# Print the result to the console
Set of 100-day date ranges generated between two points using the Arrow library

Now we simply iterate through the different date ranges and request the data for each one from Fitbit:

all_data = []
for date_range in date_ranges:
  print(f"Requesting data for {date_range[0]} to {date_range[1]}.")
  url = f"{client.API_ENDPOINT}/1.2/user/-/sleep/date/{date_range[0].year}-{date_range[0].month:02}-{date_range[0].day:02}/{date_range[1].year}-{date_range[1].month:02}-{date_range[1].day:02}.json"
  range_data = client.make_request(url)

Convert data to a Pandas DataFrame

The final step is to combine all of the data together into a manageable and easily visualisable Pandas DataFrame.

sleep_summaries = []
# Iterate through all data and create a list of dictionaries of results:
for data in all_data:
  for sleep in data["sleep"]:
    # For simplicity, ignoring "naps" and going for only "stage" data
    if sleep["isMainSleep"] and sleep["type"] == "stages":

# Convert new dictionary format to DataFrame
sleep_data = pd.DataFrame(sleep_summaries)
# Sort by date and view first rows
sleep_data.sort_values("date", inplace=True)
sleep_data.reset_index(drop=True, inplace=True)
Final Pandas Dataframe of 2 years of sleep data downloaded from the Fitbit API. This dataset is made up primarily of the per-day summaries provided by the API responses directly.

We can add some useful columns to help with visualisation now:

# It's useful for grouping to get the "date" from every timestamp
sleep_data["date"] = pd.to_datetime(sleep_data["date"])
# Also add a boolean column for weekend detection
sleep_data["is_weekend"] = sleep_data["date"].dt.weekday > 4

Visualise with Pandas and Matplotlib!

Finally for the fun bit! For this post, I’ve used the built-in integration between Matplotlib and Pandas to generate some summary graphics of my sleep data:

Sleep distributions

A two-layer histogram of time asleep vs total time in bed is a nice visualisation of “bed efficiency” that you are achieving!

# Sleep distribution

# add some nice axis labels:
ax = plt.gca()
plt.grid("minor", linestyle=":")
plt.title("Sleeping Hours")
An overlay of two histograms of total sleep minutes vs total time in bed using Fitbit data. The overlay effect is created using the alpha parameter in Matplotlib. In my own sleeping habits, there’s a relatively consistent 50-60minute difference between my total time in bed and my total time asleep.

Sleep efficiency

Along with the histogram, a scatter plot is an effective method to compare total bed duration vs total sleep duration, and adding a perfect linear fit line shows how far off “100% efficiency” my sleep is. I didn’t fit a line to the data, but the graph suggests an approximate 50 minute X-intercept lost each night to “awake in bed” moments (!), in agreement with our histogram above.

# Plot a scatter plot directly from Pandas
# Add a perfect 1:1 line for comparison
ax = plt.gca()
x = np.linspace(*ax.get_xlim())
ax.plot(x,x, linestyle="--")
A scatter plot of each night asleep over two years, showing the relationship between my total time in bed and my total time asleep. This data was downloaded from the Fitbit API using Python.

Sleep patterns over time

Show a longer term pattern of sleep over a year or two duration to see general trends. I was surprised to see how consistent sleep has been over the years, with the odd extra long or extra short sleep that, upon investigation, I could attribute to particularly early flights or restful holidays!

# Sleep makeup - calculate data to plot
plot_data = sleep_data.\
  [["deep_minutes", "light_minutes", "rem_minutes", "wake_minutes"]]

# Matplotlib doesn't natively support stacked bars, so some messing here:
df = plot_data
fig, ax = plt.subplots(figsize=(30,7), constrained_layout=True)

bottom = 0
for c in df.columns:, df[c], bottom=bottom, width=1, label=c)

# Set a date axis for the x-axis allows nicer tickmarks.

# Show a subset of data for clarity on the website:
plt.xlim(pd.to_datetime("2018-10-01"), pd.to_datetime("2020-10-01"))
Longer term sleep habits can be exposed by creating a non-summarised view of sleep durations over time. Unfortunately, the stacked bar chart is non-trivial to create in Matplotlib; mdates is also used to create the datetime based x-axis labels.

Go Further

You can use the above method to also retrieve and analysis food, activity, and heartrate data, all of which would make fantastic subject matter for further plots and exploration.

If you get stuck, there’s further blog posts and resources on the subject that are worth referencing:

Notify of

Inline Feedbacks
View all comments

I tried to follow this tutorial but I keep getting an error keyError: date when I Sort by date and view first rows
sleep_data.sort_values(“date”, inplace=True)

Please could you help me by telling me how to get the sleep data with .sort_values(“date”)? I have been at it for weeks and still cant figure it out.

There’s some error in your code.
In the first section
import arrow. #Arrow is a really useful date time helper library.
should not have a . after arrow.

Print the result to the console


#should be

in the sleep distributions you forgot the code to display the plot.

you need to add

Also the sleep efficiency section you forgot to import numpy so you need to add code.
import numpy as np
to the top of your code.

Other than that cool tutorial. I will find it helpful in my project.