joelberkeley / spidr

Accelerated machine learning with dependent types
Apache License 2.0
72 stars 4 forks source link

Mention lack of implicit broadcasting in nuisances tutorial #415

Open joelberkeley opened 1 month ago

joelberkeley commented 1 month ago

Explain that you can't e.g. add [1, 2] to [[1, 2], [3, 4]], unlike in numpy, because it would mar type inference. Explain that we currently also don't intelligently remove any broadcast/reshape/expand used to make this work, and we don't know if ML compilers will do this for you.