ai.djl.pytorch.engine.PtNDArray.gammaln - Not implemented yet.

[fredrikmoller]

Hi!

I am interested in running a DeepARNetwork on my MacBook Pro, with the M1-chip.

When running TrainTimeSeries from /examples/training, using IntelliJ, with the defaultEngine = PyTorch, this is the result:

Exception in thread "main" java.lang.UnsupportedOperationException: Not implemented yet.
	at ai.djl.pytorch.engine.PtNDArray.gammaln(PtNDArray.java:834)
	at ai.djl.timeseries.distribution.NegativeBinomial.logProb(NegativeBinomial.java:47)
	at ai.djl.timeseries.distribution.DistributionLoss.evaluate(DistributionLoss.java:54)
	at ai.djl.training.EasyTrain.trainSplit(EasyTrain.java:124)
	at ai.djl.training.EasyTrain.trainBatch(EasyTrain.java:110)
	at ai.djl.training.EasyTrain.fit(EasyTrain.java:58)
	at ai.djl.examples.training.TrainTimeSeries.runExample(TrainTimeSeries.java:132)
	at ai.djl.examples.training.TrainTimeSeries.main(TrainTimeSeries.java:78)
	Suppressed: java.lang.NullPointerException: Cannot invoke "java.lang.Float.floatValue()" because the return value of "ai.djl.training.TrainingResult.getValidateEvaluation(String)" is null
		at ai.djl.examples.training.TrainTimeSeries.lambda$setupTrainingConfig$0(TrainTimeSeries.java:209)
		at ai.djl.training.listener.SaveModelTrainingListener.saveModel(SaveModelTrainingListener.java:151)
		at ai.djl.training.listener.SaveModelTrainingListener.onTrainingEnd(SaveModelTrainingListener.java:90)
		at ai.djl.training.Trainer.lambda$close$2(Trainer.java:331)
		at java.base/java.util.ArrayList.forEach(ArrayList.java:1511)
		at ai.djl.training.Trainer.notifyListeners(Trainer.java:285)
		at ai.djl.training.Trainer.close(Trainer.java:331)
		at ai.djl.examples.training.TrainTimeSeries.runExample(TrainTimeSeries.java:105)
		... 1 more

What am I doing wrong?

//Fredrik

[zachgk]

It isn't a mistake on your end. The DJL function NDArray.gammaln(...) has not been implemented by the PyTorch engine yet. I would say you could use the MXNet engine which does implement it, except only PyTorch works on M1. If you are interested in contributing, you could add the pytorch operator to DJL. It is fairly straightforward as you have to add the function in the pytorch-native JNI calling the pytorch native gammaln and then just call it from the DJL PyTorch engine

[fredrikmoller]

Zack,

Thx for the reply.

I'm afraid I don't have the skill to add the gammaln-function but it it good to know the setup!

//Fredrik

[moberc]

Hi,
I have hit the same problem when trying to get TrainTimeSeries.java to work on Windows using pyTorch.
Does this mean that I have to switch to MXNet? Isn't that a discontinued framework? I have a hard time to get it to install on Windows.
This is my first project with djl, I need to get models to run on Windows and linux from java. I have worked with dl4j before and i did not have any of these problems. I am confused about the direction. Any advice much appreciated.
Thanks,
Margaret

[frankfliu]

Not all operators are implemented in PyTorch. Unlike MXNet, we have to implement PyTorch JNI for every operator. Do you mind contributing to DJL to add JNI for gammaln?

[moberc]

Hi, I do not have much time for this project. In fact I am so desperate that I am considering just invoking python and running my script. Is MXNext a better platform for integration with java? Thanks, Margaret

…

On Sun, Nov 5, 2023 at 1:50 PM Frank Liu ***@***.***> wrote: Not all operators are implemented in PyTorch. Unlike MXNet, we have to implement PyTorch JNI for every operator. Do you mind contributing to DJL to add JNI for gammaln? — Reply to this email directly, view it on GitHub <#2781 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AINBZF4XMQMY74EZD5PE3GDYC7GWJAVCNFSM6AAAAAA4Z5IXZOVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TIOBQHA4TO> . You are receiving this because you commented.Message ID: ***@***.***>

-- <https://smart-trade.net> *Linkedin* <https://www.linkedin.com/company/smart-trade-technologies> | *Twitter* <https://twitter.com/smarttrade_tech> | *Facebook* <https://www.facebook.com/smartTradeTechnologies/> _The content of this e-mail (including any attachments) is strictly confidential and is for the sole use of the intended addressee(s). If you are not the intended recipient of this e-mail please notify the sender immediately and delete this e-mail from your system. Any disclosure, copying, dissemination or use of its content (including any attachments) is strictly prohibited. _ _ _