EMBL Deep Learning course exercises and materials

Schedule:

• Webinar 1: May 8 11:00 - 12:00 CET
• Webinar 2: May 15 11:00 - 12:00 CET
• Webinar 3: May 22 11:00 - 12:00 CET

Set-up:

We will use on-premise EMBL resources for the course. You can also use google colab to run any of the notebooks.

BAND

We will use BAND desktops, cloud based virtual desktops provided by EMBL to work on the exercises for Webinar 2 & 3 and during the course itself. These BAND desktops are open resources provided by EMBL as part of EOSCLife. For the duration of the course they will be reserved for the course participants.

Using BAND

• Go to band.embl.de and click LOGIN.
• Use your Google account to log in. If you login for the first time, you will need to wait a few minutes to obtain a confirmation e-mail. Check out the BAND user guide or the BAND video tutorial for more information.
• This will open the Launch Desktops page
  • Select a suitable configuration in Desktop configuration, see example in the screenshot below.
  • Click LAUNCH and then yes in the prompt
  • Click GO TO DESKTOP in Running desktops
• This will open the virtual desktop in a new tab.
  • Note that for some users starting the first virtual desktop fail and display a CONNECTION ERROR. If that happens for you, just terminate the desktop and start a new one; that should fix the issue.
• You can now use this desktop to use pre-installed software or install new software
• In BAND you have write access to your directories in /home and on /scratch. /home has very limited space, therefore you should store all your data on /scratch.
  You can create your directory on /scratch like this:
  mkdir /scratch/your_name
  Data that you save on /scratch is persistent between sessions and will be stored for 3 months.
• The environment for the webinar exercises is already preinstalled via jupyterLab, you can access it via Applications->Programming->DL Course
• This will open jupyter lab. You can test it e.g. with the notebooks at https://github.com/constantinpape/dl-teaching-resources

More information

• Users can install software or download data into their directory on /scratch, which persists in between sessions.
• The required software and environments can be pre-installed (see example for DL Course environment above) so course participants don't need to set up everything on their own.
• Traning data or other large data-sets can be shared via the EMBL S3 data storage, which can be accessed via Applications->Data Access->Mount EMBL S3 Bucket

Downloading the exercises to BAND

• Open the terminal (e.g. via small terminal system in the top toolbar)
• Go to your directory on /scratch: cd /scratch/your_name
• clone this repository via git clone https://github.com/kreshuklab/teaching-dl-course-2023.git
• Now you can navigate to the folders with exercises in the jupyter environment to run the exercises
• To update the repository, e.g. to download new exercises, open the terminal again and
  • Go into the directory via cd teaching-dl-course-2023.git (or some other filepath if you did not directly clone it to your home)
  • update the repository via git pull origin main

Enabling copy paste in BAND

In many browsers copy-and-paste between your system and BAND does not work by default. You can however copy-and-paste into an intermediate menu and then copy-and-paste into/from band (copy to band explained below, pasting from BAND similarly):

• copy text on your computer
• go to the BAND desktop, press ctrl+shift+alt
• this will open a menu; now paste the copied text into the menu (ctrl+v)
• copy the text in the menu (ctrl+c)
• close the menu again via ctrl+shift+alt
• now you can copy the text into BAND

If you need to copy-and-paste a lot of text and the above procedure is incovenient for you, you can also try to activate direct copy-and-paste in your browser:

• For Chrome users:
  • Open the Google Chrome browser.
  • Type the following text in the address bar: chrome://flags#shared-clipboard-receiver.
  • Select Enabled from the drop-down list next to the Enable receiver device to handle the shared clipboard feature option.
  • Similarly, enable the flag chrome://flags#shared-clipboard-ui named Enable shared clipboard feature signals to be handled.
  • Finally, enable the flag Sync Clipboard service accessible by the address chrome://flags#sync-clipboard-service
  • Restart Chrome
• For Firefox users:
  • Open a new firefox tab
  • In the URL field, type about:config and press enter to open the config page.
  • On the config page, search "Clipboard"
  • Set the following properties to true:
    dom.events.testing.asyncClipboard
dom.events.asyncClipboard.readText
dom.events.asyncClipboard.clipboardItem
  • Delete firefox cookies
  • Restart browser

Setting up an environment on your own machine

We also provide python environments that are compatible with the course material:

• environments/pytorch contains the pytorch environment files (one for setting up an environment with GPU support, the other for setting up a CPU-only environment)
• environments/keras contains the keras environment file

You can install one of these environments with conda:

conda env create -f </path/to/<ENV-FILE-NAME>.yaml>

and then activate it via

conda activate <ENV-NAME>

If you are not familiar with conda, please refer to the conda tutorials.

Uploading data to BAND

There is no option to upload/download data between host computer and BAND desktop, so to upload data you will need to use some cloud storage, such as Google drive.

In the Applications -> Data Access you can find pre-installed Rclone, otherwise use internet browser.