Master contains a new cool feature you might appreciate. Broadcasting support! 
It was implemented with the help of two distinct types `RowVector` and 
`ColVector`. Any compatible matrix can be casted to these i.e: 
`ColVector(ones(5, 1))` and when doing arithmetic with Matrices, will be 
broadcasted to the correct dimensions, so: `zeros(5, 10) + ColVector(ones(5, 
1)) == ones(5, 10)`. This way has the advantage of being explicit as code like 
that is type checked, so no magic's going on. And also works well with `sink` 
optimizations. Previously I tried another two different ways, (1. a separate 
vector object 2. mimicking other tensor libs), and both failed to work as 
nicely with the rest of the library. So try it out and let me know what you 
think. v1.4 will be released next week.

Reply via email to