lejon / TSne.jl

Julia port of L.J.P. van der Maaten and G.E. Hintons T-SNE visualisation technique.
Other
143 stars 25 forks source link

TSne or Tsne or tsne or ...? #8

Closed alyst closed 7 years ago

alyst commented 7 years ago

Before the package is registered (#3), maybe it's worth to revisit the package name. The official name of the method is "t-distributed stochastic neighbor embedding", or t-SNE. The "t" is small, because it comes from "Student t-distribution".

The alternative names:

lejon commented 7 years ago

Yes, tsne would probably be the best I guess. I'm not entirely clear on all the details in getting it registered. Would the GitHub repo name need to be changed? Other than that I guess it is the module name? Is that all?

lejon commented 7 years ago

From a quick read I guess the correct thing would be to rename the GitHub repo to "tsne" as well. From what I understand this should not cause too much problems for people that have forked the repo and all followers etc seems to be redirected.

alyst commented 7 years ago

Yes, the name of the repo should match. Maybe you can ask @tkelman whether "tsne" is a feasible name for a package.

tkelman commented 7 years ago

A lowercase first letter would be discouraged by the module naming conventions. kNN is a very old and deprecated package, and the only exception.

Acronyms are also discouraged.

alyst commented 7 years ago

@tkelman so the only valid option is "Tsne" or even "TStochasticNeighborEmbedding"?

tkelman commented 7 years ago

I'd personally prefer something like the latter, more explicit and discoverable that way. Tab completion makes long names not such a big deal. Could shorten Stochastic to Stoch or similar maybe.

alyst commented 7 years ago

@lejon I think your package deserves better visibility and is quite stable to be registered in the package repository in its current state. I can help you with registering in METADATA.jl. The only change that only you, as the owner of the repository, can do is to rename it according to Julia conventions. IIUC, TStochasticNeighborEmbedding.jl is the variant that most of the people had agreed with (anticipating potential extensions by another distance metrics it could also be called just NeighbourEmbedding.jl or StochasticNeighbourEmbedding.jl).

tkelman commented 7 years ago

May also want to consider whether you still need to support Julia 0.4, given 0.6 is around the corner - might be worth setting the minimum in REQUIRE to 0.5 before registering, if you don't really need 0.4 support any more

alyst commented 7 years ago

I can make a PR upgrading it to Julia 0.5+ prior to registering.

juliohm commented 7 years ago

Please register this awesome package as TStochNeighborEmbedding.jl :+1:

Also, please remove the dependencies on RDatasets and MNIST. Is anyone working on this? I can submit a PR today if you want.

lejon commented 7 years ago

Done.