Hi Jochen,
Thank you for your help!!
I took a different approach to exploit DeepLearning4j for GroovyLab, and
it seems to work.
Specifically, the current build.gradle script can produce a fat jar,
with the DeepLearning4j libraries, with
gradle fatJar
From this "fat" GroovyLab, DeepLearning4j libraries seem to work well
(the Linux64GroovyLabDL4j.sh script runs this version)!
.. perhaps, GroovyLab+DeepLearning4j can become an interesting tool for
machine learning practitioners!
Thank you!
Stergios
On 04/12/2019 10:01 PM, Jochen Theodorou wrote:
On 12.04.19 11:15, sterg wrote:
Hi all,
I tried to use in GroovyLab ?(https://github.com/sterglee/GroovyLab) the
ND4j Java scientific library
(https://github.com/deeplearning4j/deeplearning4j) , by placing the
corresponding .jar file at the classpath of GroovyShell.
Unfortunately, it has problem to initialize the backends (i.e. native
code, e.g. OpenBLAS, Intell MKL, NVIDIA CUDA etc),
and code such as
x= org.nd4j.linalg.factory.Nd4j.rand(9,9)
fails.
It fails how and why?
With Java9's JShell such code works and it is possible to work with the
ND4j in a scripting MATLAB like style.
But also by using the JShell's API from GroovyLab >
I have the same problem as with GroovyShell, i.e. backend initialization
failure.
Also with GroovyConsole the backend isn't properly initialized.
I assume it is a classloader setup problem. Without knowing details it
is going to be difficult to answer something really useful though. If it
is about native libraries it can also be that those are not found. Then
it would be important to know how they are looked-up and such
bye Jochen
