In this talk, I review some developments in my research group at MIT regarding taking decisions directly from data. Starting with my paper with Nathan Kallus: ``From predictive to prescriptive analytics’’, first written in 2014 that presents a framework from extending predictive Machine Algorithms of considerable generality to prescriptive ones for two stage problems, we discuss a number of exiting new developments: With Chris Mc Cord, in the paper ``From Predictions to Prescriptions in Multistage Optimization Problems’ written in 2017, we extend the earlier framework to multistage problems and also prove rates of convergence. With Bart van Parys in the paper, ``Bootstrap Robust Prescriptive Analytics’’ written in 2017, we provide an approach to make our prescriptive approaches immune to overfitting phemonema. With Nihal Koduri in the paper, ``Data-Driven Optimization: A kernel regression approach’’, written in 2018 we provide non-parametric methods that outperform earlier approaches for two stage problems.