Keras Progress Messenger

Connect your Keras training to Discord and get notified after every epoch. Credits to @mfloto for the inspiration.

Example Image

Installation

You can use docker-compose to run the notifier. You can also build the docker image yourself. For the docker-compose, you need to edit the variables.env file the following content:

variable	content
DISCORD_WEBHOOK	The webhook url of your discord channel
PORT	The port the Bot will be listening. If you want to change this, you also need to change it in the docker-compose.yaml

After that, you can run docker-compose up -d.

Building from scratch

If you want to build the docker image yourself, you first need to build the project:

cargo build --release

Then you can build the docker image:

docker build . -t keras-training-notifier:latest

After that, you can run the docker image:

docker run -d -p PORT:PORT --env-file variables.env keras-training-notifier:latest

Usage

To use the notifier, you need to add the following code to your Keras training:

import tensorflow as tf
#...
server_callback = tf.keras.callbacks.RemoteMonitor(
    root="http://localhost:9000",
    path="",
    field="data",
    headers=None,
    send_as_json=True,
)
#...
model.fit(x, y, epochs=5, batch_size=32, validation_data=(x_test, y_test), callbacks=[server_callback])

To get a full sample have a look at the keras.py file.