Deep learning training progress window freezes

I am essentially implementing the Matlab example for image claissification on the MNIST dataset (https://www.mathworks.com/help/deeplearning/ug/create-simple-deep-learning-network-for-classification.html), the only major change being that I am reading the images in as an array rather than a data store.

My problem is that about 80% of the time the training progress window stops updating. If I press the stop early button, it updates one last time and then sits there stuck on whatever iteration it made it to. It rarely makes it to the 'Reached final iteration' stage, which means it does not return, which means that I lose all my training progress. I get no error messages, no feedback on what has gone wrong.

Update: I left the progress window open last night and this morning it had finally finished, with this message:

Warning: Graphics timeout occurred. To share details of this issue with MathWorks technical support, please include that this is

an unresponsive graphics client with your service request.

> In nnet.internal.cnn.ui/TrainingPlotPresenterWithDialog/updatePlotForLastIteration (line 128)

In nnet.internal.cnn.ui/CLITrainingPlotter/updatePlotForLastIteration (line 62)

In nnet.internal.cnn.util/TrainingPlotReporter/finish (line 79)

In nnet.internal.cnn.util/VectorReporter/computeAndReport (line 64)

In nnet.internal.cnn.util/VectorReporter/finish (line 32)

In nnet.internal.cnn/Trainer/train (line 155)

In nnet.internal.cnn.trainNetwork.doTrainNetwork (line 91)

In trainNetwork (line 181)

Exception "java.lang.ClassNotFoundException: com/intellij/codeInsight/editorActions/FoldingData"while constructing DataFlavor for: application/x-java-jvm-local-objectref; class=com.intellij.codeInsight.editorActions.FoldingData

Exception "java.lang.ClassNotFoundException: com/intellij/codeInsight/editorActions/FoldingData"while constructing DataFlavor for: application/x-java-jvm-local-objectref; class=com.intellij.codeInsight.editorActions.FoldingData

Warning: Graphics timeout occurred. To share details of this issue with MathWorks technical support, please include that this is

an unresponsive graphics client with your service request.

> In nnet.internal.cnn.ui/TrainingPlotPresenterWithDialog/showPostTrainingStage (line 140)

In nnet.internal.cnn.ui/CLITrainingPlotter/showPostTrainingStage (line 66)

In nnet.internal.cnn.util/TrainingPlotReporter/reportFinalIteration (line 93)

In nnet.internal.cnn.util/VectorReporter/computeAndReport (line 64)

In nnet.internal.cnn.util/VectorReporter/reportFinalIteration (line 57)

In nnet.internal.cnn/Trainer/doFinalize (line 275)

In nnet.internal.cnn/Trainer/finalizeNetwork (line 166)

In nnet.internal.cnn.trainNetwork.doTrainNetwork (line 94)

In trainNetwork (line 181)

Warning: Graphics timeout occurred. To share details of this issue with MathWorks technical support, please include that this is

an unresponsive graphics client with your service request.

> In nnet.internal.cnn.ui/TrainingPlotPresenterWithDialog/showPostTrainingStage (line 140)

In nnet.internal.cnn.ui/CLITrainingPlotter/showPostTrainingStage (line 66)

In nnet.internal.cnn.util/TrainingPlotReporter/reportFinalIteration (line 93)

In nnet.internal.cnn.util/VectorReporter/computeAndReport (line 64)

In nnet.internal.cnn.util/VectorReporter/reportFinalIteration (line 57)

In nnet.internal.cnn/Trainer/doFinalize (line 275)

In nnet.internal.cnn/Trainer/finalizeNetwork (line 166)

In nnet.internal.cnn.trainNetwork.doTrainNetwork (line 94)

In trainNetwork (line 181)

Answer 1

[Divya Gaddipati]

This is a known issue and the developers are working on it.

Alternatively, launching MATLAB from the "Windows Command Prompt" with the "-softwareopenglmesa" flag can fix the issue.

[Tynan Stevens]

This worked! A bit of a workaround, but good to know that a fix is in the works.