xorjane / laffinjulia Goto Github PK

Fix this!

I may need to add an explanation of type stability in week 1. Otherwise, students may run into issues with mutating functions that try to update, for example Vectors of Ints to Vectors of Floats during an axpy operation.

... rather than as matrices with single elements. This issue is related to #2.

Consider the index-free implementation of dot using FLAME. In the Matlab/Python implementation of FLAME, slicing and dicing results in chi_1 and psi_1 being represented as matrices. This is consistent with alpha being represented as a matrix, rather than a number, given the update

alpha += chi_1 + psi_1

I've chosen to represent alpha as a Number (consistent with my choice in this version to represent rank 1 objects as Vectors). For consistency, should I represent chi_1 and psi_1 as Numbers as well? I need to check if slicing and dicing ever results in chi_1 and psi_2 having multiple elements (I think this can happen but I haven't tested that case). If so, I may want to treat alpha as a matrix.

Currently I specify using .laff and using .flame, in addition to specifying the origin of functions from either module, e.g. as laff.copy!. This is redundant. I'd prefer to keep the namespace specifications and drop the using statements.

... to remove conflations.

Update function names to match Matlab notes accordingly. For example, in week 4 Tmvmult_un_unb_var2 in the Python version seems to be the same as Trmvp_un_unb_var2.

The version of LAFF in Matlab conflates rank 1 with rank 2 objects: vectors are consistently represented matrices with a single row or column. In my initial prototype of a translation to Julia, I'm representing rank 1 objects with Vectors instead. This seems more faithful to the way things are done in Julia, but may cause undesirable inconsistencies with the current offering of LAFF.

Two previous versions of the course exist: one in Matlab (current) and one in Python (2014). Translating to Julia from Python within Jupyter notebooks is more straightforward, and so that's how I've begun. However, as there are some inconsistencies between the Matlab and Python versions, this approach creates inconsistencies between my translation and the current version of the course. For example, some functions written as mutating functions in Python are non-mutating in the Matlab versions (for example, axpy in week 1). In the future I will likely want to resolve inconsistencies between the Matlab and Julia translations.