I’ll address each point -
-
It’s definitely possible to run Hail 0.2, there’s a description from a Databricks solutions architect here: Hail on Databricks with Spark Cluster. It’s possible to make this even easier, as well.
-
You are of course welcome to use whichever version you want, but we are going to stop supporting 0.1 (with either bugfixes or forum/chat help) pretty soon. The “stability” of the 0.1 release refers solely to the interface – it is a contract that we won’t push back-incompatible changes. Every commit is tested before it’s deployed, and testing practices are certainly not any worse in the 0.2 beta version than in 0.1. An 0.2 stable release is also coming, and there probably won’t be huge interface changes before that happens.
-
This isn’t a concern, you can specify multiple Kryo registrators. These classes function to make some serialized objects more lightweight by obviating the need to store class names as strings, but rather as integer references to registered classes. There should be no incompatibility issues presented by using several.
-
(BGEN) This is a good reason to think about using a Hail native format, yes. Hail can compute minimal representations of variants, but cannot realign against the reference genome right now. We have a liftover function, which is only available in 0.2.