Data Science on a Team

6 (more) tips on providing quality tailored data science services

Posted by Marcos Sponton 2 years, 7 months ago Comments

A few days ago we wrote about providing professional customer oriented data science services. For this blog post we want to focus on the team that provides those services.

  1. SHARE WITH THE TEAM: Are you working with a others? Basic descriptive statistics of the object of study should be shared (and explained, if needed) constantly, as it provides baseline of common knowledge that will make the rest of the project go smoother.
  2. DO NOT TAG ALONE: Do you need to tag data? Tagging is not a trivial task. You need to define clear tagging criteria and it’s always good to take the team members opinion into consideration. If possible, you might want to bring external talent to help you in this matter.
  3. DO NOT REINVENT THE WHEEL: If you think that 3rd party applications might help you provide a quick approximation of the final results you’re looking for, spend some time on them. Not only this will help the current project, but it’ll give you a better perspective on the next ones. Try to balance the time spent researching the theoretical aspects of the problem and trying out different tools that might provide a stopgap solution. Remember that even if a paper says that something is possible, there might not yet be a tool to prove that statement.
  4. ASSUME HIGH LEVELS OF UNCERTAINTY AND SHARE THEM WITH THE TEAM: Managing a Data Science project is different than a traditional Software development one. There are many agile approaches that have a lot of R&D but you always need to take into account higher levels of uncertainty and the client needs to know this. There is a constant trade-off between information discovery tasks and management of resources and deadlines.
  5. IT IS OK TO BUILD FLIMSY SOFTWARE: This is a constant source of clash within Machinalis. There’s only one way to make software and that is to make it right. You don’t cut corners, and you remain committed to quality all the way to release. Usually this isn’t the case on Data Science project. Of course it applies if you’re trying to build a tool (I would say it applies even more in that case), but if the goal is just to extract some information from the data, these rules need to be relaxed. Once you have a better understanding of the field of study, you then might spend additional time strengthening the solution.
  6. REMEMBER THE DATA/USER DILEMMA: Do not fall in love with the data if the users are suggesting otherwise. Do not follow the user’s advice if the data contradicts them. Remember that you might be extracting information that might contradict the user’s knowledge of the subject matter.

The last time we invited to you think about the service you’re providing, keeping the focus on the concrete needs of the clients (and their solutions) without getting enthralled by the beauty of working with data. This time, we want you to take a look at your team, your resources and processes and think on how they can help you build concrete solutions that take uncertainty out of providing a data science service.

Thanks to Agustin Barto for his cooperation to write this article.

Previous / Next posts