Urgent to priority covariance issue

Pandora-IsoMemo / ReSources

Food Reconstruction Using Isotopic Transferred Signals (ReSources)

https://isomemoapp.com/app/resources

GNU General Public License v3.0

0 stars 1 forks source link

Urgent to priority covariance issue #65

Open isomemo opened 1 year ago

isomemo commented 1 year ago

@mgross @arunge

Please give top priority to solving this over any other task!

Check this file: model test correlation.zip

Two problems

1) The proxies for the sources are correlated. I tried both repeated measurements and direct values entry options but in both cases, I get an error message that the number of proxy dimensions is wrong

2) Option missing to assign var-covar values for each source. Assuming that this may also be missing when using the repeated measurements option.

isomemo commented 1 year ago

@mgross @arunge

Problem 1 is solved (had lower/upper case difference in naming proxies).

arunge commented 1 year ago

@mgross Could you help with the 2nd point here, or give more details on what exactly should be added (e.g. screenshots)? If the model must be adjusted for this, I could first implement the UI such that you only would have to update the model afterwards. (If it's only a simple change, please just tell me where the new inputs must be added.)

isomemo commented 1 year ago

@mgross You can wait on this! I will write a fix (not only to the reported issue) using pseudo code. Should have this ready tomorrow.

mgross commented 1 year ago

@isomemo: I don't understand the second issue, should this be implemented again? A long time ago it was implemented that way (correlation between sources), but we changed it to correlation between proxies because of your request

"MVN for table sources.

This should work as in FRUITS 3.0. Currently a matrix is shown for each proxy. This should be removed.
Whereas in the specification (and in FRUITS) a single table is shown listing all proxies simultaneously. I prefer your implementation but this was likely the cause for the confusion when it came to implement the MVNs.
The variance-covariance to be shown should include simultaneously all columns displayed in each table within Sources (e.g. for the Five Source data example the elements of the variance-covariance matrix would be: Carbon-Protein, Carbon-Energy, Nitrogen-Protein) "

isomemo commented 1 year ago

@mgross I did not do a good job of explaining it. It is not the correlation between sources. It is the correlation among proxies but by source instead of a common matrix for all sources as per the mathematical descriptions and as in FRUITS. This is why I mentioned each table within Sources (meaning the table for each source). I am preparing a file with a detailed pseudo-code description.