Hello Everyone.
You all might be familiar with FluxML. However at this stage
I think it is very experimental and there might be alternatives
to achieving the following purposes for the code found HERE:
I have found MLJ.jl, but wonder if you all have some better
packages in mind, perhaps - ScikitLearn.jl?
Thanks,
In my opinion MLJ.jl is the best, though you might find it is missing a few models (and in those cases I believe you can plug ScikitLearn.jl into it). One of the best things about MLJ.jl is that you can plug all the hyperparameters into some object along with your model so that the hyperparameters are not treated as special, making life vastly easier.
Also, flux is really for neural networks specifically, MLJ is a framework, so they're not really the same thing. I believe you can plug Flux models into MLJ pretty easily now.
Expanding Man
Thank you very much! Might you have
a couple of minutes to help me work this
problem set out?
I have a made much effort, but am not
really making too much progress. Might
be more simple for you.
I suggest you just post your questions and see who can help out (I will try to help if it's something I know how to solve and I have time). I suggest you post in #machine-learning , or maybe #helpdesk (published) .
Expanding Man
Alright, I will post there. But thank you for the
tips. Found the works from Dr. Blaom on GH.
Was going to attempt to work through those
4 bullet points.
The FluxML example instructions, do not all
work on my machine with the packages I have
which include the ones they imported.
Expanding Man
Essentially, I want to adapt their approach to
MLJ.jl , it seems a better fit for the data they
are importing and the basic asks I am working
through.
07:23
Hello Everyone.
You all might be familiar with FluxML. However at this stage
I think it is very experimental and there might be alternatives
to achieving the following purposes for the code found HERE
1. Creating Prediction Model
2. Develop Training Graph
3. Obtain Sensitivity Parameter
4. Predict best performance
I have found MLJ.jl, but wonder if you all have some better
packages in mind, perhaps - ScikitLearn.jl?
Thanks,
Hello Everyone.
You all might be familiar with FluxML. However at this stage
I think it is very experimental and there might be alternatives
to achieving the following purposes for the code found HERE
I have found MLJ.jl, but wonder if you all have some better
packages in mind, perhaps - ScikitLearn.jl?
Thanks,
There are alternatives such as KNet.jl and Avalon.jl. Also I think you can take a look at TensorFlow.jl and Torch.jl if you are coming from Python.
I think you already asked the question in other streams.
Júlio Hoffimann [
Yes -- I did, was suggested that I ask in
more than one stream. Is that not good
practice?
Júlio Hoffimann
Thank you. It seems that these packages are more performant
than FluxML in some categories.
It is fine, just a heads up because it is not very common to see the same question in multiple streams :smile:
Note that MLJ is not an alternative to Flux, but interoperates with it via https://github.com/alan-turing-institute/MLJ.jl. Think of it like if you could use PyTorch models from Scikit-Learn!
@zulip-admins is it possible for us to merge this, https://julialang.zulipchat.com/#narrow/stream/238249-machine-learning/topic/Alternatives.20to.20FluxML and https://julialang.zulipchat.com/#narrow/stream/137791-general/topic/Alternatives.20to.20FluxML?
In the meantime, I will copy over my comment from another thread:
Note that MLJ is not an alternative to Flux, but interoperates with it via https://github.com/alan-turing-institute/MLJ.jl. Think of it like if you could use PyTorch models from Scikit-Learn!
QuBit said:
However at this stageI think it is very experimental and there might be alternatives
I'd also be curious to know where you get this impression from. My understanding is that Flux has been used in production at various organizations for at least a couple of years now. We take backwards compat very seriously, and, per semver, do not release breaking changes lightly.
Can do. Do you want the merged thread to be here, or in #machine-learning or in #general ?
Here, please. I'd like to have this indexable for other curious folks to find :)
This topic was moved here from #machine-learning > Alternatives to FluxML by Mason Protter
This topic was moved here from #general > Alternatives to FluxML by Mason Protter
Knet.jl and Avalon.jl both look like cool pacakges that I'd enjoy knowing how to use. I have hardly even used Flux.jl much myself though because I just don't to that much with neural nets
Brian Chen
My statements drawn from the table found HERE
Ah, that is talking about the integration between MLJ and Flux (MLJFlux.jl), and not Flux itself.
Brenhin Keller said:
Knet.jl and Avalon.jl both look like cool pacakges that I'd enjoy knowing how to use. I have hardly even used Flux.jl much myself though because I just don't to that much with neural nets
Yes, they're both very compelling and deserve more recognition!
QuBit has marked this topic as resolved.
QuBit has marked this topic as unresolved.
QuBit has marked this topic as resolved.
QuBit has marked this topic as unresolved.
Last updated: Nov 06 2024 at 04:40 UTC