Don't marry data on your first date

Saturday night. 11 PM on the coolest pub. You are just about to pay the bill and leave, but then she appears. You like her, she looks really attractive to you. She stares at you and there seems to be something going on. Then, she approaches you and says:

Hi, my name is Jessica. I'm 33 and single. Would you marry me?


Of course, this situation seems ridiculous. You've just met each other some seconds before and there's a long term proposition on the table.

However, is this that uncommon?

Think about your latest data science job hunt. You were scrolling your Linkedin feed trying not to be fooled by these data analyst offers when you suddenly saw a startup which raised 10MM last week. They're growing their data science team.

And yes, your role will also involve machine learning modeling. Let's go.

You go through the interview process and they want you on the team. You accept. You start your marriage with this company next month.

At the beginning, everything is new for you and so there're plenty of ideas on your head on how machine learning could help your new company. But you mainly have one big data set from which you should develop a machine model for X task.

You start with the exploratory data analysis and it seems that you'll be able to do everything with it. You have hundreds of ideas on the feature engineering side. Knowing the size of the dataset, you'll even try neural networks. Additionally, this will be the model which delivers most value to the company. So you'll be entirely dedicated to it.

That sounds amazing... for the first three months as maximum.

Then day 91 with the same dataset comes. No more feature engineering ideas. All the ETL pipeline has already been coded. You find some bugs and fix them. Then retrain models. After that, data distribution has changed. Retrain again. You get worse results now. More bugs on the deployment phase.

At this point, data is no longer attractive to you. You almost know it by heart. The 1% improvement you can achieve on it doesn't seem to be enough challenging to you compared with the initial situation.

You were hired for developing and deploying this model, but then what? Many marriages end up because there's a lack of attraction and daily routine kills passion.

When looking for a data science job, think about the long term. This sexy dataset can be something exciting for the first weeks of your romance. But then what? If you want to be satisfied long term, think about what'll happen next.