Add health dataset to the FL (incomplete)

Brief explanation of changes made:

Added new argument to Server and Client side called useHealthDataset, to be set as true if the health dataset is to be used
Xray model file is called xraymodel.zip, must also be in the workdir
The dataset is to be stored in the workdir under the name train_batch.bin and test_batch
Xray dataset exists in the form of a binary file, the file was converted from an image dataset using a script
The dataset is originally not in binary form and must be converted using the script
Changed BaseCifar10Loader and ICifar10Loader to BaseDatasetLoader and IDatasetLoader respectively
Overall, functions used to load the cifar 10 dataset are reused to load the xray dataset, hence the imageSize, imageHeight etc. have been modularized
Added a new class called NewChestXrayDSIterator, similar to NewCifar10DSIterator

Currently, the issue takes place on the client's side during the training stage. The specific error is below:

Exception in thread "Thread-4" java.lang.IllegalStateException: Invalid input, does not match configuration: expected [minibatch, numChannels=10, inputHeight=7, inputWidth=7]  but got input array of shape [128, 10, 26, 26]
    at org.deeplearning4j.nn.conf.preprocessor.CnnToFeedForwardPreProcessor.preProcess(CnnToFeedForwardPreProcessor.java:98)
    at org.deeplearning4j.nn.multilayer.MultiLayerNetwork.computeGradientAndScore(MultiLayerNetwork.java:2794)
    at org.deeplearning4j.nn.multilayer.MultiLayerNetwork.computeGradientAndScore(MultiLayerNetwork.java:2742)
    at org.deeplearning4j.optimize.solvers.BaseOptimizer.gradientAndScore(BaseOptimizer.java:174)
    at org.deeplearning4j.optimize.solvers.StochasticGradientDescent.optimize(StochasticGradientDescent.java:61)
    at org.deeplearning4j.optimize.Solver.optimize(Solver.java:52)
    at org.deeplearning4j.nn.multilayer.MultiLayerNetwork.fitHelper(MultiLayerNetwork.java:1753)
    at org.deeplearning4j.nn.multilayer.MultiLayerNetwork.fit(MultiLayerNetwork.java:1674)
    at org.deeplearning4j.nn.multilayer.MultiLayerNetwork.fit(MultiLayerNetwork.java:1661)
    at com.bnnthang.fltestbed.commonutils.clients.Cifar10TrainingWorker.run(Cifar10TrainingWorker.java:64)
java.lang.NullPointerException
    at com.bnnthang.fltestbed.commonutils.clients.BaseClientOperations.handleReport(BaseClientOperations.java:98)
    at com.bnnthang.fltestbed.commonutils.clients.BaseClient.coordinate(BaseClient.java:121)
    at com.bnnthang.fltestbed.commonutils.clients.BaseClient.serve(BaseClient.java:84)
    at com.bnnthang.fltestbed.commonutils.clients.BaseClient.run(BaseClient.java:55)

The model seems to work correctly when used in outside of the FL setting. So far, I have verified that the image received by the clients' side is indeed as required. The issue could be in how the model is loaded. It may also have to with how the dataset is used.

gnahtb / RealFL

Add health dataset to the FL (incomplete) #22