edit

Install Extra Dependencies

Floydhub's environments come with many common deep learning and machine learning packages and dependencies preinstalled. Examples of pre-installed packages include numpy, scipy, OpenCV, OpenAI Gym, SpaCy, etc.

If you need additional or custom packages, you can install them before running your job.

Installing Python dependencies

If your code needs additional Python packages at run time, you can add them to a special file named floyd_requirements.txt. If you want to install non-Python packages, please see here.

It is similar to Python's requirements.txt file and should be present in the same directory from where you issue the floyd run command. This is a special file that will be read before your job is started and the packages listed here will be installed before running your job.

floyd_requirements.txt

Here is an floyd_requirements.txt example file:

redis
tqdm==4.11.2

When this file is present in the project's root directory, any job that is run inside this project will have the redis and tqdm (version 4.11.2) packages installed and available at runtime.

Notes

  • Only Python packages: This will only install Python packages available in PyPi. Please ensure that the package you are trying to install is available.
  • One package per line: Ensure that you have only one package per line in floyd_requirements.txt
  • Installing specific versions: You can install specific versions of packages using the <package>==<version> notation. For example, an entry tqdm will install the latest version of the package, but tqdm==4.11.2 will force install that specific version.

Installing Non-Python Dependencies

You might want to install non-Python packages or other packages that have custom installation steps. If you are using a Jupyter Notebook, you can follow these steps to install arbitrary packages interactively.

If you are running a script using the floyd run "<command>" command, you can do one of the following:

  1. Include the installation steps in the <command>

    For example, to install OpenAI Universe before running your actual script train.py:

    $ floyd run "git clone https://github.com/openai/universe.git && cd universe && pip install -e . && python train.py"

    This will clone the Universe git repo, install it and then execute pythontrain.py.

  2. Create an installation script:

    Including the setup instructions in the run command can get unwieldy very soon. An alternative would be to create a bash script with the sequence of setup commands (say, setup.sh) and then execute this bash script as part of your floyd run.

    Here's an example of what setup.sh might look like:

    #!/bin/bash

git clone https://github.com/openai/universe.git
cd universe
pip install -e .

    Execute the setup bash script in your floyd run command before your actual job:

    $ floyd run "bash setup.sh && python train.py"

Note

Your job will run in a Linux environment. You can use the APT package manager to install dependencies using apt-get.

Installing Dependencies Inside Jupyter Notebook

You can install packages (Python or otherwise) interactively inside Jupyter Notebooks. To execute a non-Python command inside a Notebook, prepend it with !.

For example, to install textblob, you can execute !pip install textblob inside your Notebook:

Install inside Jupyter Notebook

You can also use this method to install non-Python packages. For example, to install OpenAI Universe inside your Notebook, you can execute !git clone https://github.com/openai/universe.git && cd universe && pip install -e .

Install non-Python package inside Jupyter Notebook

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.