Github user njayaram2 commented on a diff in the pull request: https://github.com/apache/madlib/pull/272#discussion_r191573063 --- Diff: src/modules/convex/mlp_igd.cpp --- @@ -241,22 +238,19 @@ mlp_minibatch_transition::run(AnyType &args) { } // tuple - Matrix indVar; - Matrix depVar; + MiniBatchTuple tuple; try { // Ideally there should be no NULLs in the pre-processed input data, // but keep it in a try block in case the user has modified the // pre-processed data in any way. - indVar = args[1].getAs<MappedMatrix>(); - depVar = args[2].getAs<MappedMatrix>(); + tuple.indVar = trans(args[1].getAs<MappedMatrix>()); + tuple.depVar = trans(args[2].getAs<MappedMatrix>()); } catch (const ArrayWithNullException &e) { return args[0]; } - MiniBatchTuple tuple; + // The matrices are by default read as column-major. We will have to // transpose it to get back the matrix like how it is in the database. --- End diff -- Can we move the above two comments to where we initialize `tuple.indVar`?
---