running in exception "cannot create Tensors with a 0 dimension: {}"

[rothkoe]

Hi, .. running into java exception

2017-10-03 22:35:06,147 ERROR [Timer-Driven Process Thread-1] c.d.p.process.TensorFlowProcessor TensorFlowProcessor[id=1f395b44-8ece-3009-d186-a257844e2506] Failed to process session due to org.apache.nifi.processor.exception.ProcessException: org.apache.nifi.processor.exception.ProcessException: java.lang.IllegalArgumentException: cannot create Tensors with a 0 dimension: {}

Any idea what this causes?

using the 2.1 version

[tspannhw]

were all the parameters set? the 2.1 version has different parameters. you need to delete the old processor and drop on a new one. make sure the model directory is readable

[simonellistonball]

Note that I also changed the expected name of model files in the modeldir, and hit this as a result. Per the README, we now assume the model is called graph.pb and the labels are in label.txt

[simonellistonball]

One other note... my bad... worked in test, not properly in a proper multi-threaded nifi... I've fixed up the approach to flowFile content reading to be more robust to race conditions, and seems to work for me now in a prod setting. Please update your build from HEAD and try again.

[rothkoe]

... parameters have been adjusted

Payload from the ListFile processor to the inbound queue to tensorflow processor looks fine
FlowFile Details UUID 571ff52e-b643-4dd9-a551-f71da3757210, Filename squirrelshirt.jpg,File Size 0.00 bytes

But zero byte found in logs/nifi-app.log loading the graph.pb (below)?

2017-10-03 22:23:58,259 INFO [Timer-Driven Process Thread-10] c.d.processors.process.TensorFlowService getInception: 0 bytes /home/nifi/demo3/models/graph.pb

the nifi-bootstrapp.log shows

2017-10-04 08:52:26,772 ERROR [NiFi logging handler] org.apache.nifi.StdErr 2017-10-04 08:52:26.772827: W tensorflow/core/framework/op_def_util.cc:332] Op BatchNormWithGlobalNormalization is deprecated. It will cease to work in GraphDef version 9. Use tf.nn.batch_normalization().

It seems the installed tensorflow framework mismatches with the version of the loaded models /graph.pb.

What do you think?

[rothkoe]

clean install of tensorflow 1.3.0 and the debug message helped to fix it

2017-10-04 11:13:23,933 INFO [Timer-Driven Process Thread-6] c.d.p.process.TensorFlowProcessor TensorFlowProcessor[id=e65b15cb-015e-1000-ca5e-3d043cbb8088] read 0 bytes from incoming file

GetFile processor works ... not ListFile (!)

Thanks

[simonellistonball]

Yep, my changes meant we now read the content from the FlowFile, not from an external file, so the best way to do it would be ListFile->FetchFile->TensorFlowProcessor, or as you say GetFile-> TensorFlowProcessor