createDatasetIterator

I create the iterator in this way:

private static DataSetIterator createDatasetIterator(int batchSize, int numBatches, List<MyData> DB) {
        List<DataSet> datasetList = new ArrayList<>();
        for (int i = 0; i < numBatches; i++) {
            datasetList.add(createDataset(batchSize, i * batchSize, DB)); //It simply takes different portions of my data
        }
        ListDataSetIterator listDataSetIterator = new ListDataSetIterator(datasetList,1); //I have also tried with 2,3,4,5.. nothing changed
        while (listDataSetIterator.hasNext()) {
            DataSet next = listDataSetIterator.next();
            int i = next.numExamples();
        }*/

        return listDataSetIterator;
}


The single DataSet is created in this  way:

private static DataSet createDataset(int batchSize, int startIndex, List<MyData> examples) {
        DataSet dataset = null;
        List<INDArray> data = new ArrayList<>();
        List<INDArray> label = new ArrayList<>();
        for (int i = startIndex; i < Math.min(startIndex + batchSize, examples.size()); i++) {
            MyData example = examples.get(i);
            INDArray tempdata = convert(example.getVectorData()); //1row 40000 columns
            INDArray templabel = example.isTheSameUser() ? Nd4j.create(new double[]{1.0, 0.0}, new int[]{1, 2}) : Nd4j.create(new double[]{0.0, 1.0}, new int[]{1, 2});
            //labes:1 row 2 cols
            data.add(tempdata);
            label.add(templabel);
        }
        INDArray dataFinal = Nd4j.vstack(data);
        INDArray labelFinal = Nd4j.vstack(label);

        dataset = new DataSet(dataFinal, labelFinal);
        return dataset;
}

THis is how my model is trained:

 model.setListeners(new ScoreIterationListener(1));//, new HistogramIterationListener(1));

        for (int i = 0; i < numEpochs; i++) {
            System.out.println("Epoch " + i);

            DataSetIterator train = createDatasetIterator(batchSize, numBatchesPerEpoch, trainUserList, loadTraceDB, random);
            model.fit(train);

            //one epoch have many batches, so this is computed rarely
            log.info("Evaluate model....");
            Evaluation eval = new Evaluation(outputNum);
            INDArray output = model.output(test.getFeatureMatrix(), false);
            eval.eval(test.getLabels(), output);
            log.info(eval.stats());
            log.info("TP  " + eval.truePositives()
                    + " FP  " + eval.falsePositives()
                    + " TN  " + eval.trueNegatives()
                    + " FN " + eval.falseNegatives());
            log.info("Accuracy " + eval.accuracy()
                    + " Precision " + eval.precision()
                    + " Recall " + eval.recall()
                    + " F1 " + eval.f1());

        }
        log.info("*** Completed training ***");

        log.info(
                "****************Example finished********************");
}