DBTB INT Scott Clark r

my name is scott clark i'm co-founder and CEO of sig ops I really enjoyed all the questions I got it was a really diverse crowd of data scientists machine learning engineers and just hard core infrastructure engineers and so it was really exciting to talk to such a diverse crowd but also kind of hear all the different challenges that they had when it came to parameter tuning these machine learning models and how cigar could potentially help I think data is really cool and exciting because people have been collecting it for many years now and things have become easier and easier to kind of through internet of things or just the bigger and bigger systems that people are building you can collect a lot more data and people are starting to attack problems and more sophisticated ways using all that data and really whenever you go to do a neural network or something like that you can just leverage sometimes petabytes of data and get really interesting result I wanted the entendes and viewers to learn that when they're dealing with these sophisticated models and building things like reading boosted machines or neural networks that they don't have to hand to them themselves data scientists can provide a ton of value by coming up with interesting features and insights and really understanding the business value of the underlying model that they're building but a lot of these models come with all these two tunable knobs and levers that hyper parameters of the model and there are better and easier and faster ways to tune them than the standard approaches of grid search and random search Beijing optimization can save you a lot of time and money my advice to people wanted my data scientists is just play with data I go on kagal go on cuanto Pia and download data sets try out new models read hacker news every day and just keep exploring and playing and there's not going to be a day where you just become a data scientist but by it playing with it every day you gain insights and you end up becoming one on accident