Use dollar street dataset

[svenvanderburg]

Fixes #445

[github-actions]

🆗 Pre-flight checks passed 😃

This pull request has been checked and contains no modified workflow files or spoofing.

Results of any additional workflows will appear here when they are done.

[tobyhodges]

Suggesting a few minor changes as I read through what you have changed already. It looks very promising @svenvanderburg. Would you like me to jump in and adjust the remaining code/output blocks (e.g. changing model names) and inserting the remaining plots that you have not already updated?

[svenvanderburg]

Thanks for your great suggestions @tobyhodges! I really like the part about provenance of the dataset👍 Yes you can pick this up if you want, but keep in mind that adding the images of all the graphs is quite a hassle :) If you don't pick it up I will pick it up in the week of 15th of April. But very much appreciated if you do it! 👍

[tobyhodges]

@svenvanderburg I have pushed a few more changes. Please take a look if and when you have time.

[svenvanderburg]

@tobyhodges great changes!

While doing some more minor changes I realised that the training really takes too long for teaching. I ended up using smaller images (64 x 64 instead of 128 x 128). This speeds up training to a workable duration and reduces overfitting a bit.

You can checkout the related changes here: https://github.com/carpentries-incubator/deep-learning-intro/pull/448/files/40b7c7e28174532002751aaa2f6fe4878a2020a5..8a3e077d56c516eb8b7abdc282485b948334dc17 Note that the flow didn't change much, it's just the dimensions of the models and different results.

I also updated the setup instructions to instruct learners to download the dataset.

I feel like this is ready to go now, can you have a final look?

[tobyhodges]

Suggesting one very minor change but otherwise this is great. Thanks for all your work on it @svenvanderburg.

I think the episode is missing one or two concluding sentences at the very end, summarising what we achieved (and did not achieve) with the models trained here, lessons learned (e.g. about training data and the difficulty of image classification tasks) and perhaps suggesting some next steps the learner could take if they wanted to improve their results. But that can be added in a separate PR.