Many junior data scientists often fall into a trap of treating modeling as hammer and data as nails. They go right into modeling without understanding the problem in the first place. Well this is the case because modeling is too sexy. On top of this, many graduate programs, bootcamps, and MOOCs often overly emphasize modeling because it is sexy to teach so too. Soo listen, for new DS, learn the business, the context, and the problems to have a holistic perspective. Dont be limited by just modeling. There are multiple approaches to a problem. Sometimes modeling is not even needed or/and is wrong approach. For schools, please partner up with companies for internship so the students can be exposed to real world work. Beside the tools and techniques emphasis, please also teach students to “think business” critically. Last but not least, please have statistics 101 and linear algebra as foundational courses. Sometimes, i see that junior DS dont even know dot product or normal distribution. They can end up doing more damage to the business due to misinterpretation or wrong application of technique. That hurts the field. My intention here is not to attack junior DS and schools/bootcamps, i observe a potential improvement, I CARE and therefore speak up. #DataScience
Thanks for this useful and helpful post. It is welcomed. a data scientist should be at the heart of business decision making. So with business knowledge and statistics background would be really good
That's what i want to say after 6 months of employment. Currently, in the school or on the Kaggle competition, problem is usually pre-defined and all we need to do is to improve the model as accurate as possible. We never have big issue in figuring out what we are going to solve. Also, we invested much of our effort in understanding statistics and math behind models at school. As it's know to all, it's not very easy to make those concepts and calculations clear, which makes us believe that's something differentiate you from others. Now, we are going to companies where time is a crucial factor. Here, we are encouraged to identify problems that are worthy of being put time and energy in; we are encouraged to use package to solve modeling problem with just several clicks instead of using Numpy from scratch or understanding how detailed calculation have been done. The priority difference between academe and industry is significant and it takes time for junior data scientist like us to get adjusted. But, event if that was said, i still believe the effort i put in school (statistics and math i used to struggle to understand) will benefit me in the future .
I agree. This is what I learnt from my little experience. What most of the data science course don't teach is the importance of analysis. They do teach EDA, but again I think their emphasis is more on fancy graphs and plot rather than on analysis per say. Analysis is rather boring and pays very high dividend. What do I mean by analysis? It's all about asking questions and finding their answers. With enough experience one will start asking better questions and thus better findings.
great insights thank you
This is so so good Kevin! Modeling data without the proper context and holistic perspective is useless. I'm glad you speak openly about this kind of challenges and gave a really nice piece of helpful advice in this crapy and overloaded LinkedIn content feed. Keep up the good work!
Totally Agree with you! Modeling is just 20% subset from actual work. The real work is to analyze and understand the pattern of the data and transform those data accord to the model.?
ЗСУ, executive data scientist; senior researcher at BICA Labs
6 年exactly! function for mapping inputs to outputs can't be found without learning territory :)