Skip to content

Score Layer Sizing in Net.evalbatch function #166

Description

@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,

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions