I have left on-going tests last week of 22.
Trying to start a X-session on ubuntu 22.04 within this new version of WSL.
Done it on a sandbox distro at Run gnome-session in Ubuntu LTS
Now gnome-shell can be started from distro ubuntu-x11 by calling gnomeshell.sh
jupyter lab
setup my base_jupyter env jupyter lab in WSL
Now jupyter lab starts automatically with distro ubuntu-22.04
and can be restarted with sudo systemctl restart jupyterlab
can install and use it including sync (otherwise no interest)
get back to fastai lesson 8
opened a ticket in wsl github : WSL some icons in taskbar for linux applications are defaulting to Tux (triaged to wslg)
Working with collab filtering (to fill NaN), inspired by fastai Lesson 8 - Practical Deep Learning for Coders 2022
seems it takes 6x times longer to use collab_learner instead of directly embeddings
I need to understand why. I know why, it was just a mistake in calling collab_learner(dls, n_factors, y_range=y_range) (forgot y_range=)
I have continued my experimentations with collaborative filtering and embedding matrices. With nice results. I have played with parameters (y_range) and hyper parameters (wd). Will have to share with colleagues to get their feedback.
I have started to use collab_learner with NN. But I have to better understand what is behind.
Continue to listen to this lesson 8 (1:06). It is back to computer vision with convolution. I have less interest for these matters but I still like the way Jeremy uses Excel to illustrate all these concepts. Jeremy: the AI excel master.
This is the end of Lesson 8 - Practical Deep Learning for Coders 2022
And a perfect time to start Lesson 9 - Deep learning foundations to stable diffusion
Stephane Mallat will give his 2023 course at College de France in the following weeks.
I guess it will be a great time as always.
In Windows Terminal (and it seems OSX as well), Ctrl-C has 2 behaviours - if some text is highlighted, then it means COPY - if nothing is selected, then it is INTERRUPT
And I like this behaviour.
Unfortunately it is not the case with gnome terminal. But what I did was to update shortcut options to map Copy with Ctrl-C.
And automatically it remaps INTERRUPT with Ctrl-Shift-C.
Not as good as Windows Terminal but good enough for now.
Just opened an issue at Detached windows cannot be used (e.g. in gimp)
I had opened another one in WSLg WSL some icons in taskbar for linux applications are defaulting to Tux (when set_app_id doesn’t match with .desktop file)
2 tips I have just used in RSS feed for my blog
To highlight some text from a file from jupyter
use | grep --color=always -z 'feed: true'
!cat ../index.qmd | grep --color=always -z 'feed: true'---
title: "blog"
listing:
contents: posts
sort: "date desc"
type: default
categories: true
sort-ui: false
filter-ui: true
feed: true
page-layout: full
title-block-banner: false
---
To insert rescaled image into jupyter
Just copy paste the image in a markdown cell:
validate cell to save this image file
and turn this into html code
<div>
<img src="attachment:24349bd1-4c85-4c51-833c-808bb8c82a38.png" width="200">
</div>
discord app in wsl
for a reason (maybe due to [snap] CA Certificates from /usr/local/share/ca-certificates are not used),
discord snap cannot (ERR_CERT_AUTHORITY_INVALID) fetch data, complete its installation and cannot be used
Got an answer from https://github.com/microsoft/wslg/issues/944#issuecomment-1399107496
for sublime-text, you can rename
/var/lib/snapd/desktop/applications/sublime-text_subl.desktopto/var/lib/snapd/desktop/applications/sublime_text.desktop. Since the application reports “sublime_text” as their app_id, just rename its .desktop file to same as it’s app_id, thanks!
Not exactly what I do: steps for sublime text in WSL
I would like to follow live coding sessions from Jeremy Howard.
1st one at https://forums.fast.ai/t/live-coding-1/96649
Those are entry level, but I am sure to learn thinks from it. And maybe detect why I have kernel dying when running stable diffusion inference from my WSL.
Just started watching Live coding sessions from fastai
Just watched video #1 of Stephane Mallat 2023.
Just watched Live coding session #3 from fastai