After normalizing, it shows better plot for the fraction of riders of different user types. And then I group them as weekdays and weekends.