In this sentence: "Plot the best validation loss and accuracy versus the number of hidden neurons.". Does this mean to do a comparison plot of the performance of all the models or just the best one with the plot_history-function?
Do as mention: x-axis should be number of hidden neurons and y-axis the best validation accuracy/loss you obtained with the corresponding model (the model with the hidden layer size as indicated in the x-axis). So rather a comparison than a single learning curve (which is validation loss/accuracy vs epochs).