Artisan ln() fit suggestion

#1: Post by **mfortin** » February 19th, 2016, 7:58 am

Suggestion to Makamo...
I used the ln() function in Artisan and got a pretty good visual fit with the RoR curve for the drying and Maillard phase up to about 30 sec post 1st crack. The results in the cup was very satisfactory. I am wondering if it could be possible to add the regression score r or r2 for the fit between the ln() curve and the RoR curve from the math screen? I think it would be a nice added tool.
Regards

#2: Post by **Nunas** » February 19th, 2016, 1:52 pm

+1

First, I REALLY appreciate Artisan and all the work that has gone into it...rarely do a roast without it. But, I too would like to see some changes in how profiles can be set up on it. But, I'm not schooled in how to do it...I just know what I know. That is, I struggle making a background profile with the included tool...just can't seem to make the line go where I want it. I've often thought that it would be nice to have either or both of: a) some typical profile curve files that could be in-built or distributed with artisan. b) some formulas that could be dropped into the math part to arrive at the same end. I realise that the capability is already there in Artisan, but for those of us who are not math wizards or computer geniuses (or just getting a bit old and dottery as may be my case), some hand-holding would be nice.

Perhaps this might be something that this group could do, without piling more work on the Artisan developer (or maybe I've not spotted such work already done)? For example, if a number of us posted saved curves of roasts we like, describing the roaster it was done on and the bean used, that would be a good start. Better still, those who know how to use the math module on Artisan could post some formulas that work?

I realise this drags this thread a bit sideways, so perhaps someone who has done work on background profiles could start a new thread?

Forever hopeful (and grateful) for the collegeial nature of this group!

#3: Post by **MaKoMo** » February 19th, 2016, 6:38 pm

mfortin wrote:Suggestion to Makamo...
I used the ln() function in Artisan and got a pretty good visual fit with the RoR curve for the drying and Maillard phase up to about 30 sec post 1st crack. The results in the cup was very satisfactory. I am wondering if it could be possible to add the regression score r or r2 for the fit between the ln() curve and the RoR curve from the math screen? I think it would be a nice added tool.
Regards

Martin,

I am confused. Why a fit between the ln() curve and the RoR? I guess you ask for a measure of the fit between the ln() curve and the actual BT curve, right?

There are some general issues regarding the use of r-square as a quality measure on non-linear approximations (see eg. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2892436/) as this ln() approximation. In the case of the ln() function there is also the issue that (at least on drum roasters), the measured BT lacks behind the real bean temperature until around the point where the beans turn yellow (DRY). That first segment would dominate (negatively) such a quality measure.

Artisan provides a simple measure to compare roasts foreground and background profiles by computing the square root of the sum of squares of the residuals (ie. the delta between the foreground and background temperature) between DRY and DROP. That measure is labeled CM in Artisan and shown as part of the HTML roast report (if a background profile is available) and also as part of the statistic line below the profile graph if "Characteristics" is ticked in the Statistics dialog (menu Config >> Statistics). Note that the statistic line has two modes. A right-click on the line switches between basic roast information like date, time, beans, charge size, weight loss, color and statistics like MET, overall RoR, ETBT area, and this quality measure CM.

So I suggest to generate a profile using the ln() function result you get using the plotter as described in the blog post on Natural Roasts (http://artisan-roasterscope.blogspot.de ... oasts.html), loading this profile in the background and let Artisan compute that CM measure. This should give

February 19th, 2016, 9:23 pm

MaKoMo wrote:Martin,

I am confused. Why a fit between the ln() curve and the RoR? I guess you ask for a measure of the fit between the ln() curve and the actual BT curve, right?

Right, I am talking about a goodness of fit between BT curve and the ln()curve (sorry for the mistake). I will tried what you suggested and see.
Thank you
Martin

February 19th, 2016, 10:05 pm

Makomo,
I did copied the ln() curve in the P2 plotter and clicked on the background button. Is there anything more to be done so as to use it as a template for future roast or will it automatically show up every new session? (I have never used the template function if it exist).
Thank you for your help
Martin

#6: Post by **MaKoMo** » February 20th, 2016, 4:40 am

mfortin wrote:Makomo,
I did copied the ln() curve in the P2 plotter and clicked on the background button. Is there anything more to be done so as to use it as a template for future roast or will it automatically show up every new session? (I have never used the template function if it exist).
Thank you for your help
Martin

The documentation for this is a little bit hidden (see the recent post on the new plotter "Signals, Symbolic Assignments and the Plotter" http://artisan-roasterscope.blogspot.de ... otter.html under Notes)

The first two plots, P1 and P2, can be turned also into an additional extra Virtual Device entry by pressing the "ET/BT" button. That way, the generated data can be saved along with the current profile.

So you would first generate a profile with that ln() curve as BT:

reset Artisan to clear the profile
copy your ln() formula into the plotter field P2 (menu Tools >> Extras, Plotter tab)
press ET/BT to add the curves described by the formulas P1 (empty here) and P2 as ET and BT curves of the current (still empty) profile
store that profile, e.g. as template.alog

Now you can load your roast profile as usual and add this template.alog profile as background profile (menu Roast >> Background, press <load>. A link to this background profile template.alog is saved along your current roast profile if you save it now.

The quality of replication measurement discussed above is shown as CM=<ET-quality>/<BT-quality>. In the example below there is no ET data in neither profile, so this one is 0 (perfect fit).

Here the "quality" is a 5.8C delta on average between DRY and DROP. That feels quite large looking at this graph (at DRY the delta in this picture is 7C). However, remember that in the above picture the profiles are align w.r.t. CHARGE (default), but the quality measure calculates the deltas from the profiles aligned w.r.t. DROP, the most critical section of a roast aligned, as shown for the same profile below (menu Roast >> Background, select "DROP" and press "Align".

Hope this helps,
Marko

PS: Thanks to Casino Mocca in Budapest , Hungary who originally suggested this feature (thus the abbreviation CM;)!

February 20th, 2016, 6:45 am

Many thanks Makomo.