This version was corrected from an error in the last version where the learning rate was set incorrectly. Both models have their strengths but this looks much more suited to the original dataset. Nothing else was changed.