Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
Meinrad Recheis
@henon
by the way, check out the examples we have
Brendan Mulcahy
@BrendanMulcahy
Yeah I check them out whenever I run into trouble :)
Meinrad Recheis
@henon
if you happen to build one that would complement our examples you might want to add it
Brendan Mulcahy
@BrendanMulcahy
BTW, the examples in the readthedocs are all super outdated
Last I checked earlier this week. I could try fixing the read the docs
Is that managed in github somehow
Meinrad Recheis
@henon
I guess it is, @Oceania2018 would know
Haiping Chen
@Oceania2018
@BrendanMulcahy Just fork repo, all the docs is in the docs folder.
Brendan Mulcahy
@BrendanMulcahy
Cool, thanks!
Haiping Chen
@Oceania2018
I’m doing the TensorFlow 2.0 in my local. But no eager mode for the first stage.
Brendan Mulcahy
@BrendanMulcahy
Do you guys prefer doing using() or the with<>(...)? I find the using more C#
Haiping Chen
@Oceania2018
We all use using for v0.11.
with<> is replaced with tf_with just for internal use.
You can use all the APIs started with tf.xxx.
Brendan Mulcahy
@BrendanMulcahy
Ok, does this work correctly then:
using (var ns = tf.name_scope(name)) { ... }
Haiping Chen
@Oceania2018
Yes, works.
Brendan Mulcahy
@BrendanMulcahy
Is there any substitute for tf.nn.in_top_k(...) ?
Eli Belash
@Nucs
gen_ops.in_top_k(...)
Brendan Mulcahy
@BrendanMulcahy
Awesome, thanks!
I didnt see it when I searched in github
Eli Belash
@Nucs
I used resharper's Ctrl+N (tick include library)
Brendan Mulcahy
@BrendanMulcahy
Good idea, although I think its bound to CTRL+T by default (search everywhere)
Meinrad Recheis
@henon

Is there any substitute for tf.nn.in_top_k(...) ?

you just found something that is missing in the tf.nn. namespace.

Brendan Mulcahy
@BrendanMulcahy
Want me to submit a PR?
Meinrad Recheis
@henon
yes please
Brendan Mulcahy
@BrendanMulcahy
Looks like there are flaky unit tests or something
Meinrad Recheis
@henon
what do you mean by 'flaky' ?
not enough, I would guess
Brendan Mulcahy
@BrendanMulcahy
Looks like the tests are failing in the AppVeyor build even though I touched pretty much nothing. Flaky meaning sometimes they succeed sometimes they fail (unless they are always failing then they arent flaky haha)
IMO You guys should consider setting up a code owners file if you havent already discussed it
Meinrad Recheis
@henon
@Oceania2018 let's discuss setting up a code owners file
Иван Сердюк
@oceanfish81_twitter
@BrendanMulcahy , I think I have two people proposed AI use cases (for gaming). Both considered Unity - but other C# (C#->native code generation) might be used
Brendan Mulcahy
@BrendanMulcahy

@BrendanMulcahy , I think I have two people proposed AI use cases (for gaming). Both considered Unity - but other C# (C#->native code generation) might be used

We took this offline, thanks for reaching out

Haiping Chen
@Oceania2018
@henon @BrendanMulcahy Add a CODEOWNERS file is a great idea. This automatically requests reviews from the code owners when a pull request changes any owned files.
Brendan Mulcahy
@BrendanMulcahy
Yeah exactly, we use it at my work and like it quite a bit. Basically the owners get pinged whenever there is a PR for a file they "own" You can setup a codeowners at the root of the project if you want to get pinged for everything. I can submit an example codeowners in a PR if you want
Haiping Chen
@Oceania2018
Yes, please.
# These owners will be the default owners for everything in the repo.
*       @Oceania2018
Brendan Mulcahy
@BrendanMulcahy
Eric Erhardt
@eerhardt
@Sergio0694 - your project looks pretty cool. Was there a reason you chose GPL for the license? A lot of .NET OSS libraries are using Apache 2.0 or MIT licenses. Of course, I'm not a lawyer, but I don't think MIT or Apache licensed libraries can use a GPL licensed library. So this may limit who can/will use ComputeSharp.
Anh B. Tran
@anhtran1010
Hi guys. I am curious about what I need to get for a simple Q-learning algorithm using the SciSharp. I know that NumSharp would probably be important but if there is anyother tips you guys would like to share for a newbie, that would be great.
Иван Сердюк
@oceanfish81_twitter

Hi guys. I am curious about what I need to get for a simple Q-learning algorithm using the SciSharp. I know that NumSharp would probably be important but if there is any other tips you guys would like to share for a newbie, that would be great.

@Oceania2018 , it looks like Anh is aware of C# and is heading to rely on Python. He was mentioned among IronPython end-users, in such a context

Sergio Pedri
@Sergio0694
Hey @eerhardt - glad to hear you like the project 😊
No particular reason for GPL3, it was just kinda my go-to license as I see it used in a lot of open source softwares.
I'll need to check the other two licenses you mentioned to see what are the main differences!
Eli Belash
@Nucs

Hi guys. I am curious about what I need to get for a simple Q-learning algorithm using the SciSharp. I know that NumSharp would probably be important but if there is anyother tips you guys would like to share for a newbie, that would be great.

That would be TF.NET which also uses NumSharp.
If you need a C# OOP architecture similar to openai/gym then you can also use Gym.NET

Haiping Chen
@Oceania2018
@Sergio0694 Is the ComputeSharp only works in Windows?
Anh B. Tran
@anhtran1010
Thank you. I will try and see if it work. I think I will also create an object recognition program just for fun and see how it goes with this
Eli Belash
@Nucs
@anhtran1010, Feel free to DM me if you need help
Sergio Pedri
@Sergio0694
@Oceania2018 Unfortunately it only works on Windows, as it needs DX12 for the shaders 😕
Johnathan Ingle
@jingle1000
Does anyone here know if it is possible to convert a NumSharp NDArray to a generic .NET array?
let x = np.linspace(-10.0, 10.0, 100)
let y = x.ToArray<double>() |> Seq.map(fun x -> x*x)
The f# function above does not work
Haiping Chen
@Oceania2018
try nd.Data<float>().
Eli Belash
@Nucs
Are you getting an exception? Because if you are using NumSharp 0.20, it should.