Part 1, online study group

shahnoza · December 1, 2019, 4:27pm

Hi @Miske! One idea is to transform text files into a dataframe. The df should have 3 columns: name of image, label of an image, and whether it is from a validation set. E.g:

Then you can use ImageList.from_df(your_df, path).split_from_df(). You can read more in the docs.

Here is the sample code that worked for me:

from fastai.vision import *
path = untar_data('https://s3.amazonaws.com/fast-ai-imageclas/oxford-102-flowers'); path

#Take txt file and turn it to df, add new col
cols = ["name", "label"]
train = pd.read_csv(path/'train.txt', sep=" ", names= cols )
train ["is_valid"] = False

valid = pd.read_csv(path/'valid.txt', sep=" ", names= cols )
valid ["is_valid"] = True

#Join train and valid dataframes
train = train.append(valid)
train.sample(4)

tfms = get_transforms()
data = (ImageList.from_df(train, path)
                 .split_from_df()
                 .label_from_df()
                 .transform(tfms, size=64)
                 .databunch()
                 .normalize(imagenet_stats))
data

You will also need to create the second DataBunch for the test data.

P.S. I am also a beginner, so there might be a more elegant solution

dharanish · December 1, 2019, 5:04pm

Hi, I would like to join as well!

shahnoza · December 1, 2019, 6:32pm

#meetup_notes

We had our 4th meetup on 30th of November! More people joined thanks to @jeremy ’s tweet Thanks everyone for the meetup and there will be another one next Saturday at 3PM GMT

Project. We talked about our first Kaggle challenge: Kannada MNIST. Our biggest trouble was to configure proper DataBunch for training set. So we discussed how to turn tabular data into images and how to add pretrained models to the kernel. We are far from the top of the leaderboard, and therefore we will keep working on the challenge during the coming week.

Lessons. More beginners joined this time. So @rugan01 went through the notebook for Lesson 1, and we discussed the main learnings from it. We also talked about cloud platforms (GCP, AWS), experience with Google Colab, data augmentation and transfer learning.

Communication. We decided to post all questions related to fastai in the fastai forum and have this thread as the main communication channel. There is also an open slack channel for discussions regarding the projects and chit chats.

P.S. The meetup is informal and dedicated for sharing, learning, and keeping each other motivated. Feel free to join anytime, you will not be late, usually, people discuss/present/ask questions from the lesson they are in.

P.S.S. Currently we are using a free version of Zoom for meetups but it disconnects every 45 mins. What do you guys think, can we get unlimited calls in any other platform? Or crowdsource/rotate payment for zoom? The upgrade is 15$.

jeremy · December 2, 2019, 4:37am

Thanks for the update @shahnoza!

Miske · December 2, 2019, 12:36pm

Thanks @shahnoza it worked perfect . Thank you this solution is elegant and more than a nought for me . You own me one .

dmilush · December 2, 2019, 3:05pm

Hi all. I am also interested to join the group. I have started some time ago and just posted my first toy project at share-your-work-here-thread and the respective notebook is available at my github. Looking forward.

shahnoza · December 3, 2019, 4:08pm

Thanks @jeremy for supporting this little community! It means a lot!

shahnoza · December 3, 2019, 4:10pm

Good to know @Miske

Great work @dmilush ! It would be great if you could show your project during the next meetup

shahnoza · December 3, 2019, 10:53pm

There is NLP study group starting here

shahnoza · December 7, 2019, 2:26pm

Hello Deep Learners! Just a reminder, we are having a meetup today at 3PM GMT, that is in 35 minutes! There is a new link to join meet.google.com/hsk-jdtn-znq.

Looking forward to the meeting

Lobster · December 7, 2019, 3:12pm

@shahnoza Is everything working alright? Currently I am to get my request to join approved.

Adeeb · December 7, 2019, 3:26pm

There were technical difficulties using the google meetings. Here is the link to the zoom meeting:

shahnoza · December 7, 2019, 3:47pm

New link for today https://zoom.us/j/707007288?pwd=V01LNUg1MDI2Q2IyRkpKK3IvbjlDQT09

shahnoza · December 7, 2019, 4:27pm

The newest one: https://zoom.us/j/667960772?pwd=d2tKK1JjdzVqeGVSRGYwbmdlRXJodz09

Jirosio · December 10, 2019, 10:48pm

Are there such groups but offline?

init_27 · December 11, 2019, 12:59pm

If you don’t have one in your area (Chances are you might)-That’s a great point to start one.

That’s what I did too

shahnoza · December 11, 2019, 3:18pm

You can also check in meetup app or facebook groups. I was able to find one through meetup in my place last year.

nchukaobah · December 11, 2019, 5:36pm

What lecture are you on now? I would like to join.

shahnoza · December 11, 2019, 6:52pm

People are mostly in Lectures 1, 2, 3,4. During the meetup everyone gives a short summary about the progress they have made during the week (lecture, personal project or group project) and we discuss interesting points. So different lectures get covered. Which lecture are you doing now?

Relativist · December 12, 2019, 1:08am

I would like to join, just finished watching lecture 1 and part of lecture 2 on YouTube. My time in front of a computer is limited outside of work, so it would be nice to be in a group so I can push myself.

I just got done building a PC with a GPU (1060 GTX) so I can run some DL. Currently it has Windows 10 pro and I was going to install pycharm, Anaconda and Pytorch. I’ve got both a 512 GB and a 1TB NVMe drive. I’d like to save $ and use that system for this course if at all possible. If someone could message me about what’s a good set of instructions for that. Otherwise I’m going to assume I need to use one of the server services mentioned in the course.

My education is as a physicist but my day job is a control systems engineer. I’ve got a few specific domains I’m interested in applying DL to.