https://kotlinlang.org logo
#ai
Title
# ai
z

zaleslaw

01/28/2021, 10:23 AM
Released  0.1.1 bugfix release https://github.com/JetBrains/KotlinDL/releases/tag/v0.1.1 Main features and bug fixes: 1. Added support batch processing for predictSoftly 2. Converted getXXX functions to properties in public API (layers and model classes) 3. Made logging based on a configuration 4. Removed unnecessary copying of FloatArray to off-heap memory in 
internalPredict
 method 5. Fixed memory leak with unclosed TensorFlow tensor in predict method 6. 4 PR from the external community members (outside from JetBrains) 6th point is significant for us; we were delighted to collaborate on issues and tickets (edited) join #deeplearning if you are interested
👍 1
a

asad.awadia

02/09/2021, 5:26 PM
Can i use this to read models generated from tensor flow?
z

zaleslaw

02/09/2021, 5:35 PM
Dear @asad.awadia could you please have a look to this tutorials https://github.com/JetBrains/KotlinDL/blob/master/docs/importing_keras_model.md#supported-architectures The one of main goals is support of keras models for inference and transfer learning. Currently, VGG and other Sequential models are supported, resnet and xception is in progress
a

asad.awadia

02/09/2021, 5:36 PM
Thank you
a

Alexandre Brown

05/09/2021, 4:30 AM
Is the library reflection free? Asking because this is a hard requirement for GraalVM application (otherwise it gets tricky)
z

zaleslaw

05/09/2021, 5:48 AM
Could you share more information about that, I have no experience with GraalVM, a reflection is used in some minor places, but it could be rewritten if there is a use-case.
a

Alexandre Brown

05/09/2021, 4:48 PM
GraalVM allows AOT compilation into a native-image executable. This executable includes the application classes, classes from its dependencies, runtime library classes, and statically linked native code from JDK. GraalVM is primarily used for backends to make the container image very small and improve performance. It has partial support for reflection, it basically needs to know ahead of time what will be using reflection if anything at all. There is a way to include some meta data to help GraalVM with reflection but it can get tricky in my opinion. Here is some more info https://www.graalvm.org/reference-manual/native-image/Reflection/
z

zaleslaw

05/09/2021, 6:36 PM
@Alexandre Brown thanks, need to experiment with that
👍 1
a

Alexandre Brown

05/10/2021, 2:35 PM
@zaleslaw This might not be the best place to ask, but is there a way we can be informed of the features of KotlinDL? Something like bullet points of the feature ex: Linear regression model, Logistic regression, k-nearest neighbours, maybe the different optimizers supported as well ex: SGD, Adam etc Thanks
z

zaleslaw

05/10/2021, 4:05 PM
Great question, @Alexandre Brown. First of all, you should raise this question in the Discussion https://github.com/JetBrains/KotlinDL/discussions or create a feature request in issues https://github.com/JetBrains/KotlinDL/issues. At the second join #kotlindl channel, all info about releases will be available there. Also, KotlinRoadmap is lighted here, in the video

https://youtu.be/jCFZc97_XQU?t=1891

👍 1
a

Alexandre Brown

05/10/2021, 4:06 PM
Thanks a lot, I will make sure to watch your video as well, keep up the great work 👍
19 Views