Exception during execution of Quickstart

[tinca]

Hello,

When trying quickstart the following is thrown:

Exception in thread "main" java.lang.IllegalArgumentException: Row 4132 does not exist in WrapCols (2708,1) at mklab.JGNN.core.matrix.AccessRow.<init>(AccessRow.java:34) at mklab.JGNN.core.Matrix.accessRow(Matrix.java:590) at mklab.JGNN.core.Matrix.accessRows(Matrix.java:779) at mklab.JGNN.nn.ModelTraining.train(ModelTraining.java:98) at mklab.JGNN.nn.Model.train(Model.java:63)

I observed that on first runs executions apparently were stuck in. Later realized that it tries to download some data. Now it quickly finishes with this error, so it is possible that something broken has been cached?

What shall I do to be able to run this properly? Thanks!

[tinca]

Further info: deleting download dir does not help.

[tinca]

One more note: I had to change one line from the example, becasue of an API change (I use 1.2.0): Slice nodes = new Slice(0, nodeLabels.getRows()).shuffle(100); to Slice nodes = new Slice(nodeLabels).shuffle(100);

[maniospas]

Hello and thanks for both the interest and the report!

• The correct way to accommodate the API change is by setting Slice nodes = dataset.samples().getSlice().shuffle(100); (the way you used it, training would erroneously use the node labels in place of the node identifiers when slicing.)

• If your dataset is as large as in the example, it takes some time to train (e.g.in my machine, each epoch needs 1-2 seconds per epoch x 300 epochs = 1-2 minutes to run). Set trainer.setVerbose(true) to get a sense of progress. By the way, if you would be more comfortable with a progress bar or want extensibility to plug in your own progress tracking mechanism, open a new issue to let us know.

I updated the readme's quickstart section to reflect these changes, so you can copy-paste the full code from there (it is also similar to examples/nodeClassification/APPNP in this repository). I am also closing this issue, but if you are still having issues replicating the quickstart, feel free to reopen it.

[tinca]

Thank you for this answer too and the quick acting on this particular issue.

When I found this library I got really excited and have high hopes about its feasibility for the problems I've in mind.