Often, the process flow of many big data projects is iterative, which means a lot of back-and-forth testing new ideas, new features to include, tweaking various hyper-parameters, and so on, with a fail fast attitude. The end result of these projects is usually a model that can answer a question being posed. Notice that we didn't say accurately answer a question being posed! One pitfall of many data scientists these days is their inability to generalize a model for new data, meaning that they have overfit their data so that the model provides poor results when given new data. Accuracy is extremely task-dependent and is usually dictated by the business needs with some sensitivity analysis being done to weigh the cost-benefits of the model outcomes. However, there are a few standard accuracy measures that we will go over throughout this book so that you can compare various models to see how changes to the model impact the result.
Data science - an iterative process
H2O is constantly giving meetup talks and inviting others to give machine learning meetups around the US and Europe. Each meetup or conference slides is available on SlideShare (http://www.slideshare.com/0xdata) or YouTube. Both the sites serve as great sources of information not only about machine learning and statistics but also about distributed systems and computation. For example, one of the most interesting presentations highlights the "Top 10 pitfalls in a data scientist job" (http://www.slideshare.net/0xdata/h2o-world-top-10-data-science-pitfalls-mark-landry)
