Delete comments <a href="http://www.aptis.info/portfolio/delete-all-comments/" target="_blank" title="Help"> <span class="aui-icon aui-icon-small aui-iconfont-help aptis-delete-all-comments-title-icon">Help</span> </a>

What Scala dependencies need to be brought in to make it work with Java.

Naveen Swamy | May 09, 2018

This is one of the first things that I'm trying to figure out and at the moment...

Andrew Ayres | May 10, 2018

I don't think it is a dependency upon the Scala compiler, or more precisely, Ma...

YiZhi Liu | May 10, 2018

In general, I am hesitant to create Yet Another MXNet Language binding. I recom...

Naveen Swamy | May 10, 2018

Just FYI, we can create 'java' folder in current scala package, and write Java ...

YiZhi Liu | May 10, 2018

Please also consider using JavaCpp https://github.com/bytedeco/javacpp https://...

Naveen Swamy | May 10, 2018

Can we also add that the APIs should decouple the MXNet internals by providing ...

Naveen Swamy | May 14, 2018

Have you measured any performance when converted between for example collection...

Naveen Swamy | May 22, 2018

Not only that you'll have to generate all the Symbol/NDArray APIs not using Mac...

Naveen Swamy | May 22, 2018

You'll also have to create Module/Executor, etc, for Java – this effort is also...

Naveen Swamy | May 22, 2018

String modelSource Does this handle the case if we have multiple files? Most M...

Qing Lan | May 22, 2018

The difference between online and batch data is significant during training. Bu...

Anirudh Acharya | May 22, 2018

latency would be of prime importance

Naveen Swamy | May 23, 2018

I support 1st approach rather than writing java bindings. There is a value in f...

Roshani Nagmote | May 22, 2018

I suggest you work on an RNN example to know what their requirements are. Visi...

Naveen Swamy | May 22, 2018

List<List<float>> is not good. inputs can be arbitrary dimension. NDArray is w...

YiZhi Liu | Aug 28, 2018

If it helps, the Clojure package uses the NDArray/array method to create an NDA...

Carin Meier | Aug 28, 2018

void predict(Collection<E> modelInput) should also support Structs stored in N...

Naveen Swamy | May 22, 2018

What are limited dependencies? Is it possible to rethink Interface C++ engine ...

Sandeep Krishnamurthy | May 22, 2018

Why make totally APIs other than reuse Module? How to use NDArray?

YiZhi Liu | May 22, 2018

DL Application Specific APIs?

Naveen Swamy | May 22, 2018

Inference Classes are good to have. But things like Module are required, which ...

YiZhi Liu | Aug 28, 2018

Can ImageClassifier be subclassed from CLassifier

Anirudh Acharya | Aug 28, 2018

Is it better to have such domain specific APIs. will it not be better to have A...

Anirudh Acharya | Aug 28, 2018

It would be hard to comment/evaluate these classes without the description of t...

Anirudh Acharya | Aug 28, 2018

Have you considered JavaCPP?

Naveen Swamy | May 22, 2018

I don't think this is relevant anymore since we decided if we did the interacti...

Andrew Ayres | Aug 28, 2018

Please mention what version of Java do you plan to support in the document

Kalyanee Chendke | May 22, 2018

Add all the open questions there? May this section should be at the top until w...

Naveen Swamy | May 22, 2018

In Problems - We should add abstracting by domain as one of the goal

Sandeep Krishnamurthy | May 22, 2018

Data structures are not clear yet. Can we design the interface and module to be...

Sandeep Krishnamurthy | May 22, 2018

getPredictions ? Can the user not wrap in a future instead of calling a separa...

Naveen Swamy | May 22, 2018

We need to add Sparse support for RNN related use-cases

Naveen Swamy | May 24, 2018

Is there a plan to include examples/tutorials with Java/Scala API?

Vandana Kannan | Aug 28, 2018

Scala examples and tutorials are ready for 1.3.

Qing Lan | Aug 28, 2018

Benchmark Scripts instead of Benchmark AI...

Qing Lan | Aug 27, 2018

Fixed

Andrew Ayres | Aug 27, 2018

any reason we have to drop Java 7?

YiZhi Liu | Aug 28, 2018

No. I was erring on the conservative side here when I wrote it but I don't have...

Andrew Ayres | Aug 28, 2018

After giving it some thought and discussions changing this to 7. We can always ...

Andrew Ayres | Aug 30, 2018

Let's please also mention we want this to be thread safe (i.e. maintains correc...

Hagay Lupesko | Aug 28, 2018

Added

Andrew Ayres | Aug 30, 2018

can you quantify? mention what steps need to be taken to use it currently and w...

Naveen Swamy | Aug 28, 2018

Added

Andrew Ayres | Aug 30, 2018

Time to market seems like a big advantage - is it correct to say we can get a w...

Hagay Lupesko | Aug 28, 2018

Excellent point. Added that.

Andrew Ayres | Aug 30, 2018

would there be an impact since most of the collection conversion is turning lis...

Naveen Swamy | Aug 28, 2018

Working on measuring this now. Preliminary testing seems like this is negligibl...

Andrew Ayres | Aug 28, 2018

Can we quantify it or be more specific? what kind of overhead are we talking ab...

Hagay Lupesko | Aug 28, 2018

I did some preliminary testing and the effects seem like it will be small to ne...

Andrew Ayres | Aug 30, 2018

Edge devices are not a relevant use case for Java API, right?

Hagay Lupesko | Aug 28, 2018

Android devices are the exception which comes to mind.

Andrew Ayres | Aug 28, 2018

Yep. But the MXNet native library will be different as well. Might be worthwhil...

Hagay Lupesko | Aug 28, 2018

measure

Naveen Swamy | Aug 28, 2018

Resolving in favor of same comment above.

Andrew Ayres | Aug 30, 2018

This I think is also an advantage. It will help drive and increase support for ...

Roshani Nagmote | Aug 28, 2018

I like the half glass full attitude!

Andrew Ayres | Aug 30, 2018

I assume the Java API release will be aligned with MXNet releases, and have the...

Hagay Lupesko | Aug 28, 2018

Added.

Andrew Ayres | Aug 30, 2018

What does it mean exactly?

Hagay Lupesko | Aug 28, 2018

I think it would mean (to me) to have a built in rnn layers and cells like in t...

Carin Meier | Aug 28, 2018

What about Sparse support? What about loading models over URL?

Hagay Lupesko | Aug 28, 2018

Something that think might be valuable in the future would be Auto Grad. Right ...

Carin Meier | Aug 28, 2018

How do you plan to support this in the absence of creating Bucketing Module or ...

Naveen Swamy | Aug 28, 2018

liuyizhi says missing control-flow support which is experimental in v1.3, so we...

Naveen Swamy | Aug 28, 2018

does this also mean Java APIs will support Gluon trained models(Hybrid Blocks)

Naveen Swamy | Aug 28, 2018

what is the reasoning to choose Java8+ only?

Naveen Swamy | Aug 28, 2018

Resolving as duplicate of another comment.

Andrew Ayres | Aug 30, 2018

IndexedSeq ?

Hagay Lupesko | Aug 28, 2018

/** * Implementation of prediction routines. * * @param modelPathPrefix Path pr...

Naveen Swamy | Aug 28, 2018

Can you also add a link to the source here ?

Piyush Ghai | Aug 28, 2018

I would suggest to share actual preliminary numbers here.

Denis Davydenko | Aug 28, 2018

List <List <Float>> predict(List <List <Float>> input) Is this a blocking call...

Hagay Lupesko | Aug 28, 2018

Would the existing Scala knowledge of handling memory not be useful here ?

Piyush Ghai | Aug 28, 2018

Lessons learned from Scala would definitely speed up the implementation.

Andrew Ayres | Aug 30, 2018

What version of Java are we targeting?

Hagay Lupesko | Aug 28, 2018

Mentioned above.

Andrew Ayres | Aug 30, 2018

epoch Since these are Inference API's, we should probably move away from using...

Vamshidhar Dantu | Aug 29, 2018

Sparse is not a must for RNN support, but it is useful to have.

Hagay Lupesko | Aug 30, 2018

This is already supported in the Scala Infer APIs

Naveen Swamy | Aug 30, 2018

talking to a potential user, they are interested to know what the throughput wi...

Naveen Swamy | Aug 31, 2018

Delete comments Help