Weekly Online zoom discussion session to discuss the weeks Lesson UTC+10 Aus/NZ

Part 1 (2019) UTC+10 Aus / NZ
21

Hi.I want to join this online discussion group. I’m from a different timezone, India. I just added my name in the list!

1 Like
22

@shub.chat Thanks, I’ve received && accepted the invitation. :slight_smile:

23

Hi, my name is Alejandro and I would like to join the group. I have added my name to the study group.

24

Thanks got it now!

25

Good to meet people! :slight_smile:

I have a few more questions for people who have done a Fast AI course before:

  • What did you learn last time?
  • Why are you doing another Fast AI course? (Was the previous course specialised on a particular topic? Do you want to do a more up-to-date version of the course? Do you just simply want to do the same course again to refresh yourself? etc.)
26

Nice to meet you too!

One of the main lessons learnt last time was the ULMFiT implementation in a specific language. It is considered, in a way, equivalent to transfer learning that has been used a lot in image processing tasks, but apparently this was not utilised to the same extent in nlp fields. I am planning to get back to work on the implementation (dropped it for a while).

I am taking the course again because this time, I would like to focus more on doing side projects with the most up-to-date fastai library (I had to put a lot of time (20+ hours) previously to just understand each lesson at a basic level). Also, wanted to have a better understanding of source codes by using them in different settings.

27

Thanks @YJP, that’s interesting! What was the specific course you did last time? And what library did they use instead of fastai 1.0?

28

Apologies - I couldn’t make last night’s zoom meeting.

I hope to be able to make it next week

29

Good morning,

I did 2018 Part 2 and I am honestly not sure what version of the library it was at the time.

The current version with its documentation is amazing because it includes walk-through examples and links to the papers implemented. Very easy to navigate around on the docs as well.

30

Hi,
I have just added my name and email to the list.

Would it be possible to join the study group?

Many thanks

Todd

1 Like
31

To all using GCP .How are you transferring files from local computer to GCP instance .I previously use to use filezilla when using paper space and it use to be a breeZe I tried the command line as well as scp but it throws an error that asset not found or something similar .Any thoughts?

34

Hey Shub,

you have multiple options you could upload to gogle drive and then use wget to download the

wget https://googledrive.com/host/LARGEPUBLICFOLDERID/file.tar

but a much better option would be to use gcloud cli

gcloud compute scp ./imageurl.txt jupyter@my-fastai-instance:~/tutorials/data

If you run in an error try to set the compute zone using

gcloud config set compute/zone $ZONE
1 Like
35

@gwittchen thanks mate. but when I am trying to download the kaggle datasets using kaggle API its throwing a 401-unauth error. Not sure what permissions I need to give to storage or what I am missing

36

Solved it.If anyone faces same problem of note being able to download using kaggle API and getting a 401-unauth error most probably the kaggle.json API key is corrupted.Delete it .Download a new one and scp transfer that onto your instance the way mentioned above.

1 Like
37

That’s because you need to share the link for everyone otherwise you need to have permission to access the file.
https://support.google.com/a/answer/6033939?hl=en

Or try using the second option that worked for me.

Kaggle cli requires setup account details and each dataset requires to accept the rules first.
Try downloading the dataset using your browser and then use the kaggle cli.

38

This works for me:
scp -i ~/.ssh/google_compute_engine ~/.kaggle/kaggle.json jupyter@YOUREXTERNALIPADDRESS:~/.kaggle/

Reverse direction is just swapping the source and destination. Refer to this link

1 Like
39

I’ve added my name to the spreadsheet - not sure how much time I will get to participate!

40

Hello my fine friends. :slight_smile: Are we still on for a call in 30 mins?

Will it be the same Zoom link as last time, or must a new Zoom session be created for each occasion?

@shub.chat You took the lead last time (thank you for doing so). Will you be running it again today?

41

Hey James.The invites went out for tomorrow mate.Yeah we will use the same zoom link as last time

42

@shub.chat Thanks for the quick response. :slight_smile:

But the lecture is also happening tomorrow, yeah? And people wanted to do the call before the lecture, didn’t they?