An in-browser playground to rapidly experiment with simple neural network architectures.

• Use the simple drag-and-drop UI to define neural network architecture
• Train and test in the browser with Tensorflow.js
• Save your model and share the permalink
• Download your model as a Tensorflow.js JSON file

This codebase uses two servers: a Flask server for the backend API and a React-based frontend (which, after it is built, can be deployed by any static server).

First install homebrew.

/usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"

Then use it to install NodeJS.

brew install node

Go install all the dependencies via

npm install

Finally, go make a production build, which bundles together the frontend code into one /build folder to be served by the server.

npm run build

To serve the build, install

npm i serve
serve -s build

Or use your preferred static server.

Now you're all set for the frontend.

Setup an SQL server and create a database. Then go to config.py to setup the SQL login.

For example, for Postgres,

SQLALCHEMY_DATABASE_URI = "postgresql://{login}:{password}@{database_url}/{tablename}"

Now we install the Python dependencies:

pip install -r requirements.txt

One last thing, be sure to migrate the database (see below). Run python run.py to start the server. Then go serve -s build to serve the frontend. Things should be working now.

If the directory server/migrations does not exist, run

python migrate.py db init

and complete the following.

Every time the SQL database structure is updated in the Python code, run the following code in order to update the SQL server.

python migrate.py db migrate
python migrate.py db upgrade