@ciheching could you clarify why you think that? It’s kind of hard to tell from your description. The basics like the channels, height, width look correct.
Double check the data from the record reader datasetiterator instead of training to make sure the data being input in to the network is correct.
the input looks like this thats why im asking
I/System.out: ===========INPUT===================
I/System.out: [[0.9961],
I/System.out: [0.9961],
I/chatty: uid=10121(com.example.train) Thread-2 identical 96 lines
I/System.out: [0.9961],
I/System.out: [0.9961]]
its supposed to be a 28x28 pixel img
@ciheching Could you describe the problem a bit? That looks like the output of a neural network not an input. Does it throw a stack trace or anything? Because you’re not showing me the complete code (only showing me what you think I need to see) it’s kind of hard to figure out what the problem is.
Are you saying that when you run the exact above code that where you’re printing the dataset it outputs a single float?
Maybe what you could do is try reproducing this setup on a desktop as well just to make sure everything works the same. I’ve never seen anyone have problems like this.
I can only assume it’s your input data or something else I’m not seeing causing this. An easier to debug environment would be a nice first step.