As a starter on this platform. giving that there are several challenges, it can be tempting to jump right into them. however, I want to believe it's more of completing the challenge and taking learning from it.
It will be nice if experts and others here can share their experience with new members.
Hey @olacharles, welcome :)
There are different categories of challenge, and some are easier to start with than others. Here are some suggestions:
Computer Vision: The approach is similar for all the image recognition challenges. Zebra vs Elephant (https://zindi.africa/competitions/sbtic-animal-classification) is a nice one to get started on, and you can then adapt your code for something like the wheat rust identification challenge (https://zindi.africa/competitions/iclr-workshop-challenge-1-cgiar-computer-vision-for-crop-disease) currently running. Good if you're keen to try deep learning!
Tabular Data: Sendy (https://zindi.africa/competitions/sendy-logistics-challenge) is a nice one to start on. You can load the main training data and make a model. Then you can merge in the riders' data (good practice combining data sources) and see if that info improves your predictions. Try to create some useful new features. And finally, if you're still looking for ways to improve your score, you can look at all the discussion around other data sources and see if you can add in something new that helps your model. These are good challenges since you can start with a very basic model and then try fancier and fancier things.
Forecasting: Predicting trends into the future. Call Volume Prediction looks like a good start (https://zindi.africa/competitions/mtoto-news-childline-kenya-call-volume-prediction-challenge), and then things like the soil moisture prediction one take things up a notch with extra variables, irrigation, weather and so on (https://zindi.africa/competitions/wazihub-soil-moisture-prediction-challenge). Tricky, but interesting.
NLP: Haven't done this one, but it looks like a decent intro to text classification: https://zindi.africa/competitions/sustainable-development-goals-sdgs-text-classification-challenge. Text/NLP can be hard, but it is an interesting domain to play around with.
Then there are challenges with satellite data or audio data, which can be more challenging to work with initially. But with these, you can usually find tips to process the data into something like a tabular format (see my approach to the ICLR #2 challenge in the starter notebook discussion) or a normal computer vision task. I'd only recommend trying these if you're looking to play with new types of data - there is plenty of the 'usual' stuff to keep you busy before this :)
Awesome! Thank you @Johnowhitaker. very succinct explanation. I will surely try out your recommendations and I believe it will help other new members that will be joining the community.
Thanks a lot
I was so excited I jumped into one of the very advanced deep learning(computer vision )competitions..now am stuck :(
but with the discussion here, I hope I can start taking baby steps on here :)
Great. I was confused as well initially hence my question. but things are clearer now.
i agree with your point @olacharles, it can be quite tough getting ahead without guidance. Also, we're here to learn first; the prize money will definitely come later after putting in hours of work
That's the mindset. Learn first, price is a reward