Verification analysis tests (comparing INCC5 and 6)

[davidbrought]

I did a comparison of the shift register verification analysis methods a while ago, here is an excel sheet with all the results (these are for samples I made up and simulated with the pulser): Comparing results (INCC5v6).xlsx

This summarizes the results..

These types of analysis that look like they aren't done yet are:

-Verification using passive multiplicity analysis (weighted multiplicity) <- INCC6 results are all 0's -Verification using passive multiplicity analysis (dual energy model) <-INCC6 doesn't ask for the required input parameters -Verification using passive multiplicity analysis (solve for efficiency, M=1) <-INCC6 doesn't solve for efficiency -Verification using passive multiplicity analysis (known alpha) <-INCC6 doesn't solve for efficiency

These give good results:

-Verification using passive calibration curve, singles rate used to calculate mass (using 5 data points in Deming to produce calibration curve in INCC5) -Verification using known alpha calibration (using 5 data points in Deming to produce calibration curve in INCC5) -Verification using passive calibration curve, doubles rate used to calculate mass (using 5 data points in Deming to produce calibration curve in INCC5) -Verification using passive multiplicity analysis (conventional multiplicity) -Known M verification analysis

[davidbrought]

Two other things that I noticed doing this was that quads are only reported in INCC5 (that wouldn't affect most people, but it's probably good to include), and there seems to be a very slight difference between the QC in INCC5 and INCC6, results match much better when QC is off (I'll recheck this to make sure the QC settings were the same).

[hnordquist]

Great! I'll get to work on all of it.

On Tue, Jan 3, 2017 at 2:02 PM, davidbrought notifications@github.com wrote:

I did a comparison of the shift register verification analysis methods a while ago, here is an excel sheet with all the results (these are for samples I made up and simulated with the pulser): Comparing results (INCC5v6).xlsx https://github.com/hnordquist/INCC6/files/683215/Comparing.results.INCC5v6.xlsx

This summarizes the results..

These types of analysis that look like they aren't done yet are:

-Verification using passive multiplicity analysis (weighted multiplicity) <- INCC6 results are all 0's -Verification using passive multiplicity analysis (dual energy model) <-INCC6 doesn't ask for the required input parameters -Verification using passive multiplicity analysis (solve for efficiency, M=1) <-INCC6 doesn't solve for efficiency -Verification using passive multiplicity analysis (known alpha) <-INCC6 doesn't solve for efficiency

These give good results:

-Verification using passive calibration curve, singles rate used to calculate mass (using 5 data points in Deming to produce calibration curve in INCC5) -Verification using known alpha calibration (using 5 data points in Deming to produce calibration curve in INCC5) -Verification using passive calibration curve, doubles rate used to calculate mass (using 5 data points in Deming to produce calibration curve in INCC5) -Verification using passive multiplicity analysis (conventional multiplicity) -Known M verification analysis

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/hnordquist/INCC6/issues/138, or mute the thread https://github.com/notifications/unsubscribe-auth/AFqkL5QEU86WyavhAVeU9uHbnw_XHmoVks5rOrd9gaJpZM4LaBcP .

[davidbrought]

Martyn's comment on this and the other issue (#139) was that since these aren't used as often it'd be better to get the collar and depending how time goes either get to these or grey them out before having other people test it.

[hnordquist]

I will prioritize in that manner.

On Tue, Jan 3, 2017 at 3:47 PM, davidbrought notifications@github.com wrote:

Martyn's comment on this and the other issue (#139 https://github.com/hnordquist/INCC6/issues/139) was that since these aren't used as often it'd be better to get the collar and depending how time goes either get to these or grey them out before having other people test it.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/hnordquist/INCC6/issues/138#issuecomment-270247732, or mute the thread https://github.com/notifications/unsubscribe-auth/AFqkL5_k3fR8oly8305LxISKunzmWu1Iks5rOs_ygaJpZM4LaBcP .

[davidbrought]

One more thing.. this is for truncated multiplicity, all of the results are also 0's, but it seems like it could be something a bit more specific: truncated mult.docx

[jflongo]

Hi I have been on vacation. If you look at the readme.txt, the missing features are listed there. The list includes those 4 multiplicity flavors as well as truncated multiplicity. Sorry you had to waste time testing discover these details. I maintain the readme.txt carefully. I assumed you read it.

There is a QC bug in the outlier code I found, it involves a specific test that checks doubles, and f doubles fails, it then checks the triples. But the bug is that even if doubles passes, it will still check triples, and triples can fail when doubles pass. So this case only occurs when triples can be computed (AMSR, JSR-15, LM VSR). I have a test that shows this bug and I will detail it for you when I get back Sunday. Cheers