DBTB INT Oscar Celma r

[Music] yeah I am Oscar Selma I'm director of research at Pandora yes regal I'm really happy to be here I'm one of the nicest things that they can catch up with friends and colleagues I used to see on these conferences apart from that also to meet new people now you present something after the representation people reach out and ask questions and it's also good for connecting and get to know where people's work wait there is so cool xn deserve um I'm sure I mean I'm not specially excited about big data and a lot of data but what can you do out of the data so just data use for the sake of data I don't feel that especially exciting things more interesting what you do on top of this data of course we have like 65 billion thumb so we have a lot of data but I wouldn't say all of this is useful but of course having a lot of data it helps a lot for your machine learning algorithms but again me I'm not excited to the data per se but like how can we just leverage all this information and cooking with something cool which products can we build on top of that more than the amount of data that the company has okay from from the attendees on my dog see it's a couple of things so I just present on this dilemma on explore explore like combining things that you know when you love and how can we know also in that case Pandora play music that you my lover that discovers like this familiarity words to discovery so i present on that and i think i provide some insights about this the pros and cons of balancing this familiarity or music you love versus things that you might be now more open to discover when's the right time to play you new music that's one thing the second one that i try to emphasize like with rough AP testing with all of testing that the users like and if they likely we roll it out to everyone but will also do especially my team generates all the algorithms all the with the site which songs to play for for all these 18 million active users so before doing any a be testing on testing things on the user's i want them to be more Dukes on quality control check their algorithm may be providing great numbers like the auc or whatever it's like it sounds it looks great but then i wanted to show me the result i want to then to be more critical about what they're doing like maybe you're okay is a new song from Metallica then we program this in Adele maybe your message safe that's great but you know that's not fair that's not that's not right so in to check that the results make some common sense that we're not playing any WTF or bad music so that goes for me before launching an online test that what will bring like a pie common sense i did make its but it's doing what you expect apart from combining with that information with the numbers that maybe are really good numbers of the results look great of line you have lot of story call data you compute things they look good okay show it to me that that it works and then we move to online testing and we check it it's a user and we see how they react how to give a data scientist yeah that's interesting so might have a team of almost 20 scientist I would say they come from different computer science eee but also like astrophysics and pure math and stats so to just like my mark wanted that you fail or more like computer science field so that's no it's very broad so the first thing i would say someone that wants to transition or becomes become a data scientist probably assuming that the technical skills were there that you're good and programming and you know python whatever sparks color all that is like that's assume that that's great then i will focus on the soft skills like how do you communicate with the non-technical people you track with no pm's can you explain to her what do you mean by the DS results can they understand that can you be in a meeting with the cpu and CTO and explain very clearly what you did so understanding the problem communicating with non-technical people do some great work that they have the charms and technical skills to do that and then how to disseminate that and see that what they're solving has an impact in the company and its online just do some research for the sake of research and finding things but connecting that to the goals of the company and then presenting that too eggs eggs or non-technical people this soft quality from your maybe not as important but like crucial as well combined with the technical skills that they're supposed to have after doing a PhD and and doing all their no technical learning all the technical aspects you [Music]