Machine learning methods are playing an increasing role in materials research, from predicting properties to accelerating characterization to extracting data from text. In this talk I will give a brief overview of recent activities and opportunities for machine learning in materials. Then I will discuss some of the exciting opportunities and open challenges specifically associated with predicting materials properties. I propose that for such models to be widely adopted we need increased accessibility and improved methods for determining model uncertainties and domains. I will discuss some recent efforts from my group and collaborators to create a new environment, called Foundry, to enable easy cloud access to machine learning data and models. I will also discuss methods to determine the uncertainties and domains in typical materials prediction models and show assessment of promising approaches. For uncertainties, I will focus on ensemble and Bayesian methods, the latter in the form of Gaussian Process regression. For domains, I will focus on approaches that consider measures of distance from training data as a guide for whether a model is applicable. I hope this talk will provide a useful introduction to those not familiar with machine learning and useful guidance for those active in the field and wishing to maximize the impact of their models.