really cool trick @mgloria! Also, I liked the dark style I will be changing mine as well
Did anyone get the buttons from 02_production.ipynb to work on Colab? Here’s what I have so far: https://github.com/WittmannF/course-v4/blob/master/nbs/02_production_colab.ipynb
Not sure if it is a bug or I am doing something wrong, but when running 01_intro.ipynb on colab, the following cell is currently raising an error:
Source: https://github.com/WittmannF/course-v4/blob/master/nbs/01_intro_colab_dev.ipynb
My previous version (from last week) was running smoothly and I didn’t change anything in that part:
Source: https://github.com/WittmannF/course-v4/blob/master/nbs/01_intro_colab.ipynb
UPDATE
Was able to run that cell by removing the input argument path
:
from fastai2.tabular.all import *
path = untar_data(URLs.ADULT_SAMPLE)
dls = TabularDataLoaders.from_csv(path/'adult.csv', y_names="salary",
cat_names = ['workclass', 'education', 'marital-status', 'occupation',
'relationship', 'race'],
cont_names = ['age', 'fnlwgt', 'education-num'],
procs = [Categorify, FillMissing, Normalize])
learn = tabular_learner(dls, metrics=accuracy)
Probably it is a last minute update in TabularDataLoaders
Not presently. The only shareable workaround I’m aware of is using the google colab file uploader (code shared below) rather than the fast.ai library uploader. The code behind the notebook widgets seem quite sophisticated.
from google.colab import files
uploaded = files.upload()
You can do the same with ipywidgets enabled JupyterLab as shown here.
You can also thry the instant colab notebook version.
After some deliberation we’ve determined this doesn’t break the Terms of Service, so the directions are in the Wiki again
FYI folks, @deeplearner and @imrandude both came up with a methodology of turning your colab instances into full Jupyter Environments. This is experimental and Colab may patch this at any moment. Please use at your own discretion. The directions are in the top post (and we simplified it a bit). If there are any install issues etc ping me and I’ll work on it.
This means you can use any Jupyter widgets in it!
This sounds great! I tried the code snippets in the wiki, yet no tunnels appear in my ngrok account - even after ~5+ minutes. Do I have to wait longer?
Looking into it now, I noticed this bug too. It was working as of last night so let me investigate
For now here are the steps to run it in one go while I fix the script:
!git clone 'https://github.com/fastai/course-v4.git'
!echo Setting up server...
!curl -sL https://deb.nodesource.com/setup_10.x | sudo -E bash -
!sudo apt-get install -q -y nodejs
!pip install -q jupyter jupyterlab --upgrade
!pip install -q jupyter_contrib_nbextensions && jupyter contrib nbextension install
!pip install -q -r /content/course-v4/requirements.txt
!wget -q -c -nc https://bin.equinox.io/c/4VmDzA7iaHb/ngrok-stable-linux-amd64.zip
!unzip -qq -n ngrok-stable-linux-amd64.zip
get_ipython().system_raw(f'./ngrok authtoken {authtoken} && ./ngrok http --log=stdout 8888 > ngrok.log &')
!perl -pi -e 's|execution_count": null|execution_count": 1|g' course-v4/nbs/*ipynb
!nohup jupyter notebook --no-browser --allow-root --ip=0.0.0.0&
And then to kill the server:
!kill $(ps aux | grep './ngrok' | awk '{print $2}')
Script works now, apologies! See the updated one in the top post
Can I know, why you are doing this?
I included it in the script as there are few issues with running notebooks due to an error caused by execution count. The line basically takes care of that. If you are not facing any issues with running the notebooks, you can remove that line
Great, thank you! Can you tell me what the bug was? I couldn‘t find it…
An issue with running the bash scripts basically. I was trying to do it via curl
but I had to save the script via wget then run it with sh bash.sh
I just wanted to say that i love this little hack. Having access to a cheap/free gpu and using the default jupyter gui makes me so much more productive, including achieving this with a single function call.
Thanks for your efforts!
I concur to this.
Hi all. So after the dust has settled: there now a standard, working, best-practice notebook for setup in Colab, and if so, where can I find it?