TensorFlow doesn't work in h2o with R

Question

I have built latest h2o version (3.11.0.99999) with deepwater and I'm getting error after running h2o.deepwater with TensorFlow backend.

This is my code:

model_tf<-h2o.deepwater(x=2:ncol(train),
                        y=1,
                        backend = "tensorflow",
                        training_frame = train
                        )

This is error:

java.lang.RuntimeException: Unable to initialize the native Deep Learning backend: resource mlp_1093x1x1_32.meta not found.

java.lang.RuntimeException: Unable to initialize the native Deep Learning backend: resource mlp_1093x1x1_32.meta not found. at hex.deepwater.DeepWaterModelInfo.setupNativeBackend(DeepWaterModelInfo.java:259) at hex.deepwater.DeepWaterModelInfo.(DeepWaterModelInfo.java:206) at hex.deepwater.DeepWaterModel.(DeepWaterModel.java:227) at hex.deepwater.DeepWater$DeepWaterDriver.buildModel(DeepWater.java:131) at hex.deepwater.DeepWater$DeepWaterDriver.computeImpl(DeepWater.java:118) at hex.ModelBuilder$Driver.compute2(ModelBuilder.java:173) at hex.deepwater.DeepWater$DeepWaterDriver.compute2(DeepWater.java:111) at water.H2O$H2OCountedCompleter.compute(H2O.java:1240) at jsr166y.CountedCompleter.exec(CountedCompleter.java:468) at jsr166y.ForkJoinTask.doExec(ForkJoinTask.java:263) at jsr166y.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:974) at jsr166y.ForkJoinPool.runWorker(ForkJoinPool.java:1477) at jsr166y.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:104)

Error: java.lang.RuntimeException: Unable to initialize the native Deep Learning backend: resource mlp_1093x1x1_32.meta not found.

The same code but with "mxnet" backend is working well.

Did you also build deepwater yourself from the latest master or did you download it/have an old jar? We've implemented custom size MLP in tensorflow just recently. Any chance you could share a sample failing dataset? — Mateusz Dymczyk

Mateusz Dymczyk Mateusz Dymczyk · Accepted Answer · 2017-05-18T03:18:52

As pointed out here this functionality does work with the latest H2O-3 and DeepWater, you do need to build both from source (or use our docker image) and install the tar.gz R package.

Older DeepWater versions will not work as we were generating TF models only for certain types of MLP models.

TensorFlow doesn't work in h2o with R

1 Answers