Accuracy scores of the merger or non-merger classification versus epoch during the training phase. The first and second panels from the top are the result of the grid search performed on galaxies at low redshifts. The third and fourth panels from the top are the results of training on the entire dataset ranging from z = 0 to 3. The title of each panel indicates the parameter which the runs are grouped on. In other words, each panel shows the average scores of all runs containing the parameter values shown in the legend. From top to bottom, grid searches are grouped on the learning rate, batch size, weight decay, and number of blocks tuned. The solid lines correspond to the training set and the dashed lines correspond to the validation set.