Mercury is a perfect tool to convert Python notebook to web app and share with non-programmers.
- You define interactive widgets for your notebook with the YAML header.
- Your users can change the widgets values, execute the notebook and save result (as html file).
- You can hide your code to not scare your (non-coding) collaborators.
- Easily deploy to any server.
The YAML config is added as the first raw cell in the notebook.
The web app is generated from the notebook. Code is hidden (optional). User can change parameters, execute notebook with the Run button, and save results with the Download button.
The demos with several example notebooks are running at:
- http://mercury.mljar.com (running on AWS EC2 t3a.small instance)
- http://mercury-demo-1.herokuapp.com (running on Heroku, if dyno is sleeping and notebooks are not loaded, please refresh it and wait a little)
You can share as many notebooks as you want. There is a gallery with notebooks in the home view of the Mercury.
You need to add YAML at the beginning of the notebook to be able to run it as a web application in the Mercury. The YAML configuration should be added as a Raw cell in the notebook. It should start and end with a line containing "---". Below examples of how it should look like in the Jupyter Notebook and Jupyter Lab:
Allowed parameters in YAML config:
title- string with a title of the notebook. It is used in the app sidebar and the gallery view.author- string with a author name (optional).description- string describing the content of the notebook. It is used in the gallery view.show-code- can beTrueorFalse. Default is set toTrue. It decides if the notebook's code will be displayed or not.show-prompt- can beTrueorFalse. Default is set toTrue. If set toTrueprompt information will be displayed for each cell in the notebook.params- the parameters that will be used in the notebook. They will be displayed as interactive widgets in the sidebar. Each parameter should have an unique name that corresponds to the variable name used in the code. Read more about available widgets inparamsin the documentation.
Widgets in Mercury are text, slider, range, select, checkbox, numeric, file.
Definition of the widget (in params) starts with the widget name. It will correspond to the variable in the code. The name should be a valid Python variable.
To define the widget you need to select the input type. It can be: text, slider, range, select, checkbox, numeric, file.
For each input we need to define a label. It will be a text displayed above (or near) the widget.
You can read more about widgets in our wiki page.
You can easily create files in your notebook and allow your users to download them.
The example notebook:
- The first RAW cell.
title: My app
description: App with file download
params:
output_dir:
output: dir- The next cell should have a variable containing the directory name. The variable should be exactly the same as in YAML. This variable will have assigned a new directory name that will be created for your user during notebook execution. Please remember to define all variables that are interactive in Mercury in one cell, just after the YAML header (that's the only requirement to make it work, but is very important).
output_dir = "example_output_directory"- In the next cells, just produce files to the
output_dir:
import os
with open(os.path.join(output_dir, "my_file.txt"), "w") as fout:
fout.write("This is a test")In the Mercury application, there will be additional menu in the top with Output files button. Please click there to see your files. Each file in the directory can be downloaded.
There is an option to set a custom welcome message. Like in the example screenshot below.
The custom welcome message can be set as Markdown text (with GitHub flavour). To set custom message please create a welcome.md file and include your Markdown text there. When deploying please set the WELCOME environment variable pointing to your file. For example, in Heroku it will be WELCOME=welcome.md. The example repository with welcome message is here. The example demo showing a Data Science Portfolio is here.
If you don't set the welcome message a simple Welcome! will be displayed. We belive that setting welcome message will give you a great opportunity for customization.
You can install Mercury directly from PyPi repository with pip command:
pip install mljar-mercury
To run Mercury locally just run:
mercury runserver --runworker
The above command will run server and worker. It will serve Mercury website at http://127.0.0.1:8000. It won't display any notebooks because we didn't add any. Please stop the Mercury server (and worker) for a moment with (Ctrl+C).
Execute the following command to add a notebook to Mercury database:
mercury add <path_to_notebook>
Please start the Mercury server to see your apps (created from notebooks).
mercury runserver --runworker
The Mercury watch command is perfect when you create a new notebook and want to see what it will look like as a web app with live changes.
Please run the following command:
mercury watch <path_to_your_notebook>
You can now open the web browser at http://127.0.0.1:8000 and find your notebook. When you change something in the notebook code, markdown, or YAML configuration and save the notebook, then it will be automatically refreshed in the web browser. You can track your changes without manual refreshing of the web app.
Please use add command to add a notebook to the Mercury. It needs a notebook paath as an argument.
Example:
mercury add notebook.ipynb
Please use delete command to remove notebook from the Mercury. It needs a notebook path as an argument.
Example:
mercury delete notebook.ipynb
Please use list command to display all notebooks in the Mercury.
Example:
mercury list
Running in production is easy. We provide several tutorials on how it can be done.
- Deploy to Heroku using free dyno
- Deploy to AWS EC2 using t2.micro
- Deploy to Digital Ocean (comming soon)
- Deploy to GCP (comming soon)
- Deploy to Azure (comming soon)
The docker-compose must be run from the Mercury main directory.
Please copy .env.example file and name it .env file. Please point the NOTEBOOKS_PATH to the directory with your notebooks. All notebooks from that path will be added to the Mercury before the server start. If the requirements.txt file is available in NOTEBOOKS_PATH all packages from there will be installed.
Please remember to change the DJANGO_SUPERUSER_USERNAME and DJANGO_SUPERUSER_PASSWORD.
To generate new SECRET_KEY (recommended), you can use:
python -c 'from django.core.management.utils import get_random_secret_key; \
print(get_random_secret_key())'
Please leave SERVE_STATIC=False because in the docker-compose configuration static files are served with nginx.
The docker-compose will automatically read environment variables from .env file. To start the Mercury, please run:
docker-compose up --build
To run in detached mode (you can close the terminal) please run:
docker-compose up --build -d
To stop the containers:
docker-compose down
The Mercury project consists of three elements:
- Frontend is written in TypeScript with React+Redux
- Server is written in Python with Django
- Worker is written in Python with Celery
Each element needs a separate terminal during development.
The user interface code is in the frontend directory. Run all commands from there. Install dependencies:
yarn install
Run frontend:
yarn start
The frontend is served at http://localhost:3000.
The server code is in the mercury directory. Run all commands from there. Please set the virtual environment first:
virtualenv menv
source menv/bin/activate
pip install -r requirements.txt
Apply migrations:
python manage.py migrate
Run the server in development mode (DEBUG=True):
python manage.py runserver
The server is running at http://127.0.0.1:8000.
The worker code is in the mercury directory (in the apps/notebooks/tasks.py and apps/tasks/tasks.py files). Please activate first the virtual environment (it is using the same virtual environment as a server):
source menv/bin/activate
Run the worker:
celery -A server worker --loglevel=info -P gevent --concurrency 1 -E
Looking for dedicated support, a commercial-friendly license, and more features? The Mercury Pro is for you. Please see the details at our website.
