Australia Study Group

That would be great. :slight_smile:

4 Likes

Yes please

1 Like

Thank you @amir8 for setting this up. Was great to meet everyone and looking forward to catching up next week. :slight_smile:

2 Likes

How is everyone tracking over the last week? Maybe a little bit of a weekly update from everyone here is in order?? If we can share our learnings here it may provide a succinct summary for everyone when read together?

For my part most of my work was on the production side and not too much training on my own datasets yet. One artifact of my progress includes a working app hosted on github pages that predicts pet breeds. It’s not the prettiest or customised but it’s a base on which I can post future models.

Other learnings

  • Customising default directories on Google Colab so trained models are saved to your Google Drive rather than the transient cloud instance.
  • I made a tiny tweak to Jeremy’s Kaggle NLP notebook changing the Transformer model and broke the model trainer. Thanks to a suggestion from the forum I got the model working using a different starting point. You can see my forum post and kaggle solution to find out more.

Now it’s over to you.

“If you want to go fast, go alone. If you want to go far, go together”. African proverb

5 Likes

Hi everyone! Just wanted to check if we will be meeting up again this week, and if so, whether it will be at the same place and time (5:00 at St Lucy’s Caffe & Cucina).

1 Like

My progress this week was training an image classifier to distinguish between serif and sans-serif fonts, and discovering that it performs very poorly on data from sources different from that of the training and validation data.

Also, I figured out how to deploy my model to HF Spaces! Check it out: Font_classifier - a Hugging Face Space by skalyan91.

5 Likes

Hi Siva

Not this week. Hopefully we plan something for next week and preferably Zoom call so that more people can join.

1 Like

Hi

Anyone have the link to Lesson3? The university did not send out the link to the youtube.

Thanks

Hi Stephen,

Here is the link fast.ai Live - Lesson 3 - YouTube

Hi team,
not sure where to post my question :frowning:

I am trying to do the Pet Gradio in the paperspace and I keep on getting


NameError Traceback (most recent call last)
/tmp/ipykernel_295/1397371048.py in
3 dls = ImageDataLoaders.from_name_func(’.’,
4 get_image_files(path), valid_pct=0.2, seed=42,
----> 5 label_func=RegexLaberller(pat = r’^([^/]+)_\d+’),
6 item_tfms=Resize(224))

NameError: name ‘RegexLaberller’ is not defined

Here is my notebook

Typo on RegexLabeller - RegexLabeller

You can see a list of places to post questions here:

1 Like

Hi all, I’m a little late to the party would love to join this study group too! I’m a PhD student doing some MRI and bioinformatics on Motor Neuron Disease at UQ.

2 Likes

OH MY GOD, you have no idea how many times I reviewed my code… Thanks.
I put it on my covid brain :frowning:

1 Like

Hello all. I’d like to propose doing a project with the Australian team as this will definitely help with learning and sharing knowledge. I have little knowledge but am very keen on improving bit by bit day by day.

I propose something achievable by perhaps the 7th lesson so it’s less likely to be abandoned. If anyone has any ideas or are interested please post here.

I have some ideas but is unsure of the feasibility or how much work is involved. One idea is a Travel buddy - use google api, set a destination then questions will pop-up asking for more clarifying questions like travel duration, forms of prefered transportation, landmarks, activities etc. Then the ML will give a recommendation based on input. Perfect case would be to get recommendations based on google reviews or pair this up with the Lonely Planet api to get top recommendations for landmaks, hotels etc. This is quite ambitious so an MVP or something would also be cool.

Let me know what you guys think.

Hi Jack

Interesting idea! So let me see if I understand this:
Starting point: Point A
Destination: Point B

Use Google map API to give possible routes from A to B based on user preferred transportation.
Use Lonely Planet API combined with Google reviews to retrieve recommendations around hotels,etc.

I guess if the aim is to find out reviews with higher ranks, right? Perhaps with the use of NLP. However, reviews on Google has already ranked with stars e.g 5 stars to 1 star. So one could use these stars as an indicator of how an review would be relevant? I am just trying to get my head around the application of ML here.

Other questions are: On what data you want to train the model? What model you want to use? What accuracy metrics are relevant here?

Hi Amir,

The idea, which needs the details to be ironed out more, is to scour through the most recent reviews on venues/accomodation using sentiment analysis plus the rating system and maybe other variables like recommendation from lonely planet or another platform then give a recommendation on sights, restaurants, accommodation etc according on travel duration, transportion and maybe budget constraints then plot the trip on google maps and have a summary on daily activities, restaurants, accommodation etc. Then be able to tweak, change and remove details afterwards. Kind of like a first draft for a travel itinerary which can then be modified manually to craft a complete holiday in as little time as possible.

To be honest I haven’t thought of all the details yet but it would be nice to discuss it with whoever is interested then move forward to creating a working prototype. As long as we get a minimum viable prototype working by week 7 then I believe that’s a good way to learn plus having a deadline helps to shave away what can and cannot be achieved. It doesn’t have to be this project either, I am only giving a suggestion. Happy to hear other project ideas as well.

Hi guys

Wonder if you think to participate on hackathon competition which is the last week of June over 3 days? I think it will be an interesting session from pitching ideas to practically coding. If you are interested, we can form a team and go from there.

Also, if you have an idea, feel free to put it here and see if we can implement it. Not sure the hackaton should be around our own ideas or we are given a project?

5 Likes

I’m really interested in the hackathon. I think it will be a great way to test out some of things we have been learning on differnet problems. As for the being given a problem or our own ideas I think it can’t hurt to have an idea before we go into the hackathon.

3 Likes

I would be interested in forming a group too:) I also have a idea. Interested to know that you all think.

Motor neuron disease (MND) is a fast evolving research field. Clinicians, researchers and patients often want to stay up to date with the latest research. I’m thinking of building a model that inputs abstracts from research papers from the past 6 months and outputs a paragraph summary of the findings.

I read somewhere that someone built something similar, but for summarising a single abstract into a sentence. I want to do something similar but on a larger scale for MND.

1 Like