Score Layer Sizing in Net.evalbatch function

[winstrom]

Hi,

I am running into some array dimension mismatches in the Net.evalbatch function due to a change in the code. Currently, the master branch has:

	val scores = zeros(score_layers.length, batchSize);
	for (i <- 0 until score_layers.length) {
		scores(i,?) = score_layers(i).score;
	}

which has given the scores matrix a vector for each layer rather than a scalar for each layer.

Previously, each score layer only had a scalar score available:

	val scores = zeros(score_layers.length, 1);
	for (i <- 0 until score_layers.length) {
		scores(i) = score_layers(i).score.v;
	}

Is there a reason that this has changed? It certainly seems to be making e.g. the GLM layer break when used as an output layer.

Change happened here:
1ff00a9

Cheers,