Hi mọi người,

Có những lúc chúng ta được chia sẻ những kinh nghiệm, những lời khuyên khá là hữu ích trong học hành và nghiên cứu, hoặc chính bản thân ta đúc kết được, song đó chỉ là những mẩu lẻ tẻ bất chợt khó có thể lập thành một topic cụ thể nào. Vậy tớ lập ra topic này để mọi người cùng chia sẻ những tips & tricks kiểu vậy vào đây nhé.

Món đầu tiên, vớ được từ discussion board của môn Advanced Statistics (mà tớ phải chuyển sang audit ngay sau set bài tập đầu tiên vì nó quá xương xẩu với tớ, hic):

Hỏi:
My experience thus far with machine learning has been that there is basically an infinite universe of possible models and approaches and frameworks and such, so I've just sort of been wondering if anyone has any experiential wisdom about how, given a new problem, you can even begin to select a specific approach to try, when you decide that approach categorically won't work out and just generally how to structure the relationships and differences between all these things in my mind. Where do I start thinking about this? Thanks.

Thầy giáo trả lời:
Here are some rules of thumb that have served me well:

- Keep things simple until you have a reason to try something more complicated.
- Explore the data a lot and try to really understand your data.
- Try to write down a statistical model for your data. This serves many purposes: it forces you to think about the sources of variability in your data; it separates model building from model fitting; and for the model fitting, much is known in statistics about appropriate ways to fit models.

Mostly, though, it just takes a lot of practice.

-Matt

__________________
Happy are those who dream dreams and are ready to pay the price to make them come true
To view links or images in signatures your post count must be 10 or greater. You currently have 0 posts.