An Easy 3-Step Guide To Get You Started On Your First AI Project
We all know that the hardest part of any project is getting started. And in a field like data science where there’s massive interest yet so many ways to break into the field, that initial hurdle is higher than most.
The company I work for, bitgrit, recently launched their own AI competition platform, with the first competition prompt asking for algorithms that will increase adoption rates at animal shelters to help our collaborator, the nonprofit Save All The Pets. Compared to other competitions that we have in the works, this is a pretty easy competition, with the added benefit of being for a great cause.
But how do you get started on such a project? While I’m no data scientist, I do work with people in the field and was lucky to be able to consult them for advice on how to tackle such a project. I compiled their advice into this easy three-step guide, so whether you’re just getting started as a data scientist in general or jumping into this competition in particular, the following advice should come in handy.
1. Cleanliness is next to godliness, so for Christ’s sake, clean up your data.
“Incorrect or inconsistent data leads to false conclusions,” claims the highly-esteemed Toward Data Science journal in an article about data cleansing. It seems that in every case, starting with a solid base of uniform, quality data beats having a kick-ass algorithm.
A key to cleaning your data is that you must understand it, as well as what your goal is. In the case of the Save All The Pets competition, our goal is to predict a dog’s breed just based on its picture using a dataset of 260,000 dog photos. By combining this algorithm with demand data, Save All The Pets will be able to relocate dogs to different shelters where they are more likely to be adopted. By participating in the competition, you get to try out your data science chops, Save All The Pets gets useful AI solutions for their worthy cause, and shelter dogs get adopted. Everyone wins!
Just a quick note: to ensure that you develop the best solution possible, divide the dataset into test and train data, usually in an 80/20 split.
2. Don’t rush — be strategic about your algorithm.
Once you’ve completed the data cleansing, be strategic about what algorithm you use to solve the competition prompt. Is machine learning sufficient, or should you utilize neural networks? The great thing about our competitions is that you can choose whatever tools you like — all we look at is the final result.
If you decide to go the machine learning route, maybe give feature engineering a try. If you prefer deep learning, why not explore loss function? Also, don’t rule out trying both as multiple approaches are likely to lead to a better, more holistic result.
3. Submit your solution for instant feedback.
Once you arrive at a solution you feel confident in, submit it on our competition website. We suggest that you submit it as soon as possible, because when you click “submit,” the system automatically scores your solution. Also, you can submit up to 25 solutions by the deadline on September 22, so if you want to aim for a higher score and the $150 first-place award, it’s best to get that feedback sooner rather than later!
That’s all for this quick 3-step guide to get you started on an AI project. I hope it was helpful, and all of us here at bitgrit are looking forward to seeing your take on this competition prompt.
For more information about the dog breed prediction competition, check out our press release, and if you have any questions then you can reach out to us anytime.
Note: This competition is now closed. You can find a list of bitgrit’s other AI competitions on our website here.
