- Joined
- Dec 28, 2006
- Messages
- 162
very cool point to point graph but not very helpful. See how much error there is from cut to cut, thus, to remove the error and get the mean for the system take an average and extrapolate an actual smooth curve out of it like I did by hand and how I use to do in my engineering days? That will give you the true direction of the curve. The graph shown is an exact representation of the written data, it helps nothing since there is no average or mean curve represented.
I'm not sure what you mean by "errors" in the data. I just see noisy data with precision only to the tens place. As far as smoothing the data goes, I agree that you can filter the data, but filtering only smooths the data. It doesn't allow you to extrapolate. The way I usually do this is to do some sort of curve fit. This usually requires me to have a of model of the trend I'm matching. Since my background is in Electrical Engineering, I'm not familiar with a good model to match this data to. So I decided to try a least-squares fit using a 3rd order, 5th order and logarithmic model.
The 420HC data was interesting since it seemed to show the greatest coverage from "sharp" to "dull"
The only insight I gain from this plot is that my models don't match the data, and that it's a non-trivial task to determine a reasonable model. If you have a better model, I'd be interested in running it against the data. However, drawing in a trend by hand is too subjective in my opinion. It allows people to make the data fit any trend they wish it to fit. Lets look at the rest of the data sets.