Introduction

This repository contains a class definition called FloydhubKerasCallback that can be used to keep track of training phase metrics when deploying your deep learning model on a floydhub server using the Keras.
Metrics can later be visualized in your floydhub job panel under "Training metrics" at real time

Usage

Copy the class FloydhubKerasCallback defined on this file to your model training script

Then instantiate the class and pass the object to the 'callbacks' parameter when calling the method fit()

model = Sequential()
# ...
model.compile(...)

callback = FloydhubKerasCallback(mode='batch')
mode.fit(X_train, y_train, epochs=100, verbose=False, callbacks=[callback])

Now deploy your model on a floydhub gpu/cpu and train it. Training metrics will be shown on the floydhub jobs panel at real time

floyd init <floydhub-project-id>
floyd run --gpu --env keras "python your_script.py"

Example

example.py script defines a deep learning model to classify hand-written digits (it uses the well known MNIST image dataset)
Execute the next code to deploy and train the model on floyhub

git clone https://github.com/Vykstorm/floydhub-keras-metrics.git
cd floydhub-keras-metrics
floyd init <floydhub-project-id>
floyd run --gpu --env keras "python example.py"