Parser

Parsing is done in a separate thread from learning. The parser uses a queue in which it places the examples it reads.

Parsing is generally a big bottleneck compared to the speed of online gradient descent. That's why VW offers the option of writing the examples in its native format, in a file we call a cache.