edit

Jupyter Notebook

Jupyter or IPython Notebooks allow you to create and share documents that contain live code, equations, visualizations and explanatory text. It is great for interactive development of code. This guide will show you how to run a Jupyter notebook on Floyd.

Setup project

Clone a project which contain deep learning jupyter notebooks. See some great Tensorflow Notebook examples at floydhub/tensorflow-notebooks-examples. Then initialize a floyd project inside that.

$ git clone https://github.com/floydhub/tensorflow-notebooks-examples.git
$ cd tensorflow-notebooks-examples/3_NeuralNetworks
$ floyd init neural-networks

Run in Jupyter mode

Floyd run command has a jupyter mode. This will upload the jupyter notebooks in the current directory and start a jupyter server for you.

$ floyd run --mode jupyter
Syncing code ...
RUN ID                  NAME                          VERSION
----------------------  --------------------------  ---------
dMoDZaCcvQMyfNbgwTx9f8  floydhub/neural-networks:1          1

Path to jupyter notebook: https://www.floydhub.com:8000/EQDsTpeqB3RkjpHUgBGDyB

To view logs enter:
    floyd logs dMoDZaCcvQMyfNbgwTx9f8

You can open the link to your jupyter server (example https://www.floydhub.com:8000/EQDsTpeqB3RkjpHUgBGDyB). All the notebooks in your project should be available for you to run.

Jupyter

Selecting the environment

Jupyter notebooks run in the same environments as other jobs. You can select any environment you want along with the --env parameter of the run command. You can see the list of supported environments here.

$ floyd run --mode jupyter --env tensorflow-0.12:py2
Syncing code ...

To use Jupyter Notebook in a GPU instance, use the --gpu flag with the run command.

Additional dependencies

If you have any additional python dependencies, you can also add a floyd_requirements.txt file to the notebook directory before floyd run. The packages specified there will be installed before running the jupyter server

Passing Input data and storing Output data

Similar to regular jobs, any data passed during floyd run using the --data flag will be mounted at /input (see Using Datasets). Any data that you write to /output will be stored for you, even after you end your interactive Jupyter session (see Managing Output).

Saving Jupyter Notebooks

Jupyter notebooks are saved as run outputs. You can view and download them after the Jupyter instance is stopped. You can use the output command for this.

Stopping Jupyter Server

Once you have experimented with your code, you need to manually stop your "job". Run the stop command for this. Remember Jupyter notebooks are charged for the entire duration they are up, not just when you execute code. So make sure the stop the notebooks when you are no longer working on them.

Help make this document better

This guide, as well as the rest of our docs, are open-source and available on GitHub. We welcome your contributions.