Porting magenta-demos to C#

[dirkmcpherson]

Hey I'm about to start porting over the sketchRNN demo to c#. Before I start I was wondering whether you knew of any resources I could use a starting point or had any advice, it seems like it'll be at least a 40 hour job so any help would be appreciated.

Thanks for making these demos public, they're really interesting and cool.

[twobob]

Are you porting them in the sense of creating a cradle that calls the extant python installation and just mirroring the functionality via wrapping the response?

[dirkmcpherson]

@twobob right now I'm just porting one of the demos for use in a project I'm working on, so I'm doing the simplest thing and building in a system to read the pre-trained weights, recreate the RNN, pass in a user's interaction, and get out the result.

[hardmaru]

Hi Recently I’m also in the process of porting sketchrnn to TensorFlow.js using TypeScript which is more or less a rewrite with webgl acceleration. The newer version also uses the same json weight files (but is the decoder-only model). So you can check out this for more resources when you are porting over to C# https://github.com/hardmaru/magenta-js/tree/master/sketch That being said this is work in progress and so I won’t be able to provide much further support.

…

On Tue, Jul 31, 2018 at 1:35 AM dirkmcpherson ***@***.***> wrote: @twobob <https://github.com/twobob> right now I'm just porting one of the demos for use in a project I'm working on, so I'm doing the simplest thing and building in a system to read the pre-trained weights, recreate the RNN, pass in a user's interaction, and get out the result. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#42 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGBoHpgVTl2YvQqBciDZpfs8FtJPIvFrks5uLzXEgaJpZM4Vh1MO> .

[dirkmcpherson]

@hardmaru thanks! No worries on further support its always really helpful to have multiple examples of the same idea. Definitely helps understanding.

[dirkmcpherson]

@hardmaru hey thanks for linking your port, it definitely helped conceptualize what was going on. You don't have temperature in your sample method, just a heads up that I wasn't able to get anything that looked remotely as good as the demo without incorporating the softmax temperature stuff in the example.