Deeplearning4j: LSTM example for review sentiment analysis

Question

I am looking through the example of deeplearning 4j for classifying movie reviews according to their sentiment. ReviewExample

At line 124-142 the N-dimensional arrays are created and I am kind of unsure what is happening at these lines:

Line 132:

features.put(new INDArrayIndex[]{NDArrayIndex.point(i),
NDArrayIndex.all(), NDArrayIndex.point(j)}, vector);

I can image that .point(x) and .point(j) address the cell in the array, but what exactly does the NDArrayIndex.all() call do here?

While building the feature array is more or less ok what is happening there I get totally confused by the label mask and this lastIdx variable

Line 138 - 142

            int idx = (positive[i] ? 0 : 1);
            int lastIdx = Math.min(tokens.size(),maxLength);
            labels.putScalar(new int[]{i,idx,lastIdx-1},1.0);   //Set label: [0,1] for negative, [1,0] for positive
            labelsMask.putScalar(new int[]{i,lastIdx-1},1.0);   //Specify that an output exists at the final time step for this example

The label array itself is addressed by i, idx eg column/row that is set to 1.0 - but I don't really get how this time-step information fits in? Is this conventional that the last parameter has to mark the last entry?

Then why does the labelsMask use only i and not i, idx ?

Thanks for explanations or pointer that help to clarify some of my questions

Answer 1

It's an index per dimension. All() is an indicator (use this whole dimension). See the nd4j user guide: http://nd4j.org/userguide

As for the 1. That 1 is meant to be the class for the label there. It's a text classification problem: Take the window from the text and word vectors and have the class be predicted from that.

As for the label mask: The prediction of a neural net happens at the end of a sequence. See: http://deeplearning4j.org/usingrnns

Answer 2

write a test and you will know it.

val features = Nd4j.zeros(2, 2, 3) val toPut = Nd4j.ones(2) features.put(Array[INDArrayIndex](NDArrayIndex.point(0), NDArrayIndex.all, NDArrayIndex.point(1)), toPut)

the result is [[[0.00, 1.00, 0.00],
[0.00, 1.00, 0.00]],

[[0.00, 0.00, 0.00],
[0.00, 0.00, 0.00]]] it will put the 'toPut' vector to the features.