SiminaB
commented
4 years ago - Part of tip 2 should also now go into Tip 1, since a simpler ML model may suffice. Eg the second paragraph:
Depending on the amount and the nature of the available data, as well as the task to be performed, deep learning may not always be able to outperform conventional methods. As an illustration, Rajkomar et al. [13] found that simpler baseline models achieved performance comparable with that of DL in a number of clinical prediction tasks using electronic health records, which may be a surprise to many. Another example is provided by Koutsoukas et al., who benchmarked several traditional machine learning approaches against deep neural networks for modeling bioactivity data on moderately sized datasets [14]. The researchers found that while well tuned deep learning approaches generally tend to outperform conventional classiers, simple methods such as Naive Bayes classication tend to outperform deep learning as the noise in the dataset increases.
- I'm not a huge fan of the heritability example, because there's no citation. Like, this could be a problem, sure, but has it ever been a problem in practice? We can use reference 39 as an example instead (or another similar example). This is also related to Tip 8, since if you can't interpret the models you don't even know what possible confounders may have gotten included.
Benjamin-Lee
This is to discuss outstanding issues for Tip 1, on whether Deep Learning should be used in the first place, https://github.com/Benjamin-Lee/deep-rules/blob/master/content/03.ml-concepts.md. Related to: https://github.com/Benjamin-Lee/deep-rules/pull/241