DXJupyterLab Reference

This page is a reference for most useful operations and features in the DNAnexus JupyterLab environment.

DXJupyterLab is accessible to all users of the UK Biobank Research Analysis Platform and the Our Future Health Trusted Research Environment.

A license is required to access DXJupyterLab on the DNAnexus Platform. Contact DNAnexus Sales for more information.

Download Files from the Project to the Local Execution Environment

Bash

You can download input data from a project using dx download in a notebook cell:

%%bash
dx download input_data/reads.fastq

The %%bash keyword converts the whole cell to a magic cell which allows us to run bash code in that cell without exiting the Python kernel. See me examples of magic commands in the IPython documentation. The ! prefix to achieves the same result:

! dx download input_data/reads.fastq

Alternatively, the dx command can be executed from the terminal.

Python

To download data with Python in the notebook, you can use the download_dxfile function:

import dxpy
dxpy.download_dxfile(dxid='file-xxxx',
                     filename='unique_name.txt')

Check the dxpy helper functions for details on how to download files and folders.

Upload Data from the Session to the Project

Bash

Any files from the execution environment can be uploaded to the project using dx upload:

%%bash
dx upload Readme.ipynb

Python

To upload data using Python in the notebook, you can use the upload_local_file function:

import dxpy
dxpy.upload_local_file('variants.vcf')

Check the dxpy helper functions for details on how to upload files and folders.

Download and Upload Data to Your Local Machine

By selecting a notebook or any other file on your computer and dragging it into the DNAnexus project file browser, you can upload the files directly to the project. To download a file, right-click on it and click Download (to local computer).

You may upload and download data to the local execution environment in a similar way, that is, by dragging and dropping files to the execution file browser or by right-clicking on the files there and clicking Download.

Use the Terminal

It is useful to have a terminal provided by JupyterLab at hand, which uses bash shell by default and lets you execute shell scripts or interact with the platform via dx toolkit. For example, the following command confirms what the current project context is:

$ dx pwd
MyProject:/

Running pwd shows you that the working directory of the execution environment is /opt/notebooks. The JupyterLab server is launched from this directory, which is also the default location of the output files generated in the notebooks.

To open a terminal window, go to File > New > Terminal or open it from the Launcher (using the "Terminal" box at the bottom). To open a Launcher, select File > New Launcher.

Install Custom Packages in the Session Environment

You can install pip, conda, apt-get, and other packages in the execution environment from the notebook:

%%bash
pip install torch
pip install torchvision
conda install -c conda-forge opencv

By creating a snapshot, you can start subsequent sessions with these packages pre-installed by providing the snapshot as input.

Access Public and Private GitHub Repositories from the JupyterLab Terminal

You can access public GitHub repositories from the JupyterLab terminal using git clone command. By placing a private ssh key that's registered with your GitHub account in /root/.ssh/id_rsa you can clone private GitHub repositories using git clone and push any changes back to GitHub using git push from the JupyterLab terminal.

Below is a screenshot of a JupyterLab session with a terminal displaying a script that:

  • sets up ssh key to access a private GitHub repository and clones it,

  • clones a public repository,

  • downloads a JSON file from the DNAnexus project,

  • modifies an open-source notebook to convert the JSON file to CSV format,

  • saves the modified notebook to the private GitHub repository,

  • and uploads the results of JSON to CSV conversion back to the DNAnexus project.

This animation shows the first part of the script in action:

Run Notebooks Non-Interactively

A command can be run in the JupyterLab Docker container without starting an interactive JupyterLab server. To do that, provide the cmd input and additional input files using the in input file array. The command runs in the directory where the JupyterLab server is started and notebooks are run, that is, /opt/notebooks/. Any output files generated in this directory are uploaded to the project and returned in the out output.

The cmd input makes it possible to use a papermill tool pre-installed in the JupyterLab environment that executes notebooks non-interactively. For example, to execute all the cells in a notebook and produce an output notebook:

my_cmd="papermill notebook.ipynb output_notebook.ipynb"
dx run dxjupyterlab -icmd="$my_cmd" -iin="notebook.ipynb"

where notebook.ipynb is the input notebook to "papermill", which needs to be passed in the "in" input, and output_notebook.ipynb is the name of the output notebook, which stores the result of the cells' execution. The output is uploaded to the project at the end of the app execution.

If the snapshot parameter is specified, execution of cmd takes place in the specified Docker container. The duration argument is ignored when running the app with cmd. The app can be run from commandline with the --extra-args flag to limit the runtime, for example, dx run dxjupyterlab --extra-args '{"timeoutPolicyByExecutable": {"app-xxxx":{"\*": {"hours": 1}}}}'".

If cmd is not specified, the in parameter is ignored and the output of an app consists of an empty array.

Use newer NVIDIA GPU-accelerated software

If you are trying to use newer NVIDIA GPU-accelerated software, you may find that the NVIDIA GPU Driver kernel-mode driver NVIDIA.ko that is installed outside of the DXJupyterLab environment does not support the newer CUDA version required by your application. You can install NVIDIA Forward Compatibility packages to use the newer CUDA version required by your application by following the steps below in a DXJupyterLab terminal.

# NVIDIA-smi
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.129.06   Driver Version: 470.129.06   CUDA Version: 11.4     |
|-------------------------------+----------------------+----------------------+
// Let's upgrade CUDA 11.4 to 12.5
# apt-get update
# apt-get -y install cuda-toolkit-12-5 cuda-compat-12-5
# echo /usr/local/cuda/compat > /etc/ld.so.conf.d/NVIDIA-compat.conf
# ldconfig
# NVIDIA-smi
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.256.02   Driver Version: 470.256.02   CUDA Version: 12.5     |
|-------------------------------+----------------------+----------------------+
// CUDA 12.5 is now usable from terminal and notebooks

Session Inactivity

After 15 to 30 minutes of inactivity in the JupyterLab browser tabs, the system logs you out automatically from the JupyterLab session and displays a "Server Connection Error" message. To re-enter the JupyterLab session, reload the JupyterLab webpage and log into the platform to be redirected to the JupyterLab session.

Last updated

Was this helpful?