Ziyang Zhou

There are tons to learn from many resources. One thing I would like to add is many companies have great engineering blogs. Engineers and scientists will be sharing some of the most practical cutting-edge methods that are used heavily in industry, which can be a great resource to learn.

I rarely find pie chart useful. It’s actually also easy to misinterpret the data while you are looking at a pie chart because people are just in general better at comparing length and heights. Not saying it’s completely useless, but I guess 99% of the time you can replace a pie chart with a bar plot which is just much easier to read into. The following article may be helpful too: https://scc.ms.unimelb.edu.au/resources/data-visualisation-and-exploration/no_pie-charts#:~:text=Pies%20and%20doughnuts%20fail%20because,important)%20are%20tricky%20to%20show.

In general training an offline model (like what we did for homework) is completely different from training an online model that serves product users. Models that perform well offline may not work as expected when we move it to production, so there are many tweaks of model weights/params you will have to do to make it really useful.

There are definitely circumstances that you will have to build a model, but I don't think *every* problem needs to be solved by a model. In fact, building a model to answer questions can sometimes over-complicate the problem and solution. Also, the more fancy your model is, the more difficulty you are adding to yourself when you explain the results to your business partners. Most of the time, I tend to just find the simplest solution whenever possible that just have the right amount of information to inform decisions.

I find GPT can actually do a pretty good job already to do text summarizations for people who may not have a deep knowledge of the cutting edge NLP techniques. It's not mentioned here, but worth to note.

I don’t really think any DS should master all the tools mentioned here. Most of the time your time will just be spent on a very small subset of those, depending on what part of data science you are anchored on. For me (analytics data scientist at LI) at least 80% of my work is done with a combo of SQL, Spark, Python, and Google sheet/slides/doc.