slackbot.txt

slackbot [11:17 PM]
Basically, pick something attainable, it doesn't really matter what. Devon and Nash picked, scraped, cleaned, and wrangled 50+ GB datasets, and it took us months, (neither model worked), and it was way harder than it needed to be. At the end it was clear that the capstone isn't supposed to be ground breaking research, you don't need to solve world hunger or crack the stock market. The goal of the capstone is to just do some solid data science. And here's the part so often overlooked, the report is the whole damn show. The most common piece of advice we give is to write the report first, before you implement anything, that way you know what you're doing and what you need to achieve. It helps keep you focused, limit the amount of work you do, and results in a higher level of quality overall. Some good ideas to get you started are:     *1.* A kaggle competition, the dataset is there and the benchmark is already laid out for you, giving you time to focus on that sweet write up     *2.* If you're interested in NLP you can try sentiment analysis, lots of good write ups so you're not all on your own    *3.* If you want to work on financial data remember it's time series and that you're not going to crack the stock market, if it could be done the guys with money wouldn't stop dumping money at it.     *4.* In addition to the suggestions on Udacity, check these resources for ideas of valuable ML/AI projects to work on: https://ai-on.org/projects/ and https://openai.com/requests-for-research/ and https://blog.openai.com/requests-for-research-2/   There are lots of other options, but remember *keep it doable and do it well*