/deep_learning

Primary LanguagePythonMIT LicenseMIT

various issues on deep learning.

Issue #1: MinPy vs NumPy

comparison between minpy and numpy on RNN implementation for binary addition problem problem: it appears that minpy's autograd is much slower than hard-coded analytic formulas for gradients. here, an implementation on binary addition problem using minpy is shown to compare with Peter Roelants' implementation on pure numpy (http://peterroelants.github.io/posts/rnn_implementation_part02/)