Colab - ImportError

[dyuliu]

I launched colab notebooks and tried to run cells one by one.

But I encountered the ImportError. (1)

(2)

[sarahmish]

Thanks @dyuliu for pointing this issue.

The problem here is that you need to restart runtime after finishing the installation and run the block again, because you are uninstalling default pandas and urllib versions in colab.

# install dependencies then restart kernel and run again
! pip install orion-ml
! pip install 'urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1'

! git clone https://github.com/signals-dev/Orion.git
! mv Orion/notebooks/tulog/* .

The second part of cloning the repo and copying files is causing the issue for you (because it has not been run). I am wondering if you have an idea of a better installation mechanism than basic pip install and git clone commands. Ideally, I wish to have a persistent environment that colab always runs from but I don't think this is possible.

It's a similar issue pointed to in #153.

[sarahmish]

Main issue When you run !pip install orion-ml cell, orion installs some versions of libraries, for example pandas, that are counter to the default version in google colab. Thus, colab sends a subtle message of asking the user to restart runtime. If this task is ignored, the user will run into an error in the rest of the notebooks (as demonstrated by @dyuliu).

Discussion of possible solutions

• add an exit() command immediately after the pip install command to ensure the kernel would be reset.
• put an error message to inform users that you should be installing orion then restarting kernel.
• catch the exception caused by the import section cell, indicating to the user to restart runtime.

Considerations to keep in mind

• If colab is reminding the user, is it our responsibility to also remind them?
• Binder is a better setup for our use case, should we also support colab or keep it at the discretion of the user?

Improvements to the current setup

• Colab actually asks the user to restart runtime, but it's really hidden (buried) under the installation output, to solve this we can add -q to pip installations
• Break the cell into (1) intalllation cell (2) cloning cell.
• Include urllib installation in orion installation -- ensure this in the new release.
• Show a successful installation message to the user by displaying orion version.