I cannot find rules in CPA

cellprofiler-analyst
#1

capture.PNG1264x486 47 KB

this is how my CPA looks after training. it only shows list of features, not the exact rules! in this case, how can I identify rules to be used by CP’s FilterObject module?

instead, is it possible to classify object outside CPA and import the labelled object in CP? I mean, I have exported a list of object IDs and their features as a .csv file, so I can feed the file as an input for other ML programs like Weka or R and make an output list of object IDs and the classification result. then is it possible to import the object ‘Positive’ or ‘Negative’ based on the output list?

Cell Paint: Rules file from CPA and FlagImage module of CP
#2

You’re only getting features instead of rules because of the classifier you’re using- if you use Fast Gentle Boosting rather than Random Forest, you’ll get a list of rules instead.

instead, is it possible to classify object outside CPA and import the labelled object in CP? I mean, I have exported a list of object IDs and their features as a .csv file, so I can feed the file as an input for other ML programs like Weka or R and make an output list of object IDs and the classification result. then is it possible to import the object ‘Positive’ or ‘Negative’ based on the output list?

You can import a training set- probably the easiest thing to do is to export the one from the classifier you have now so that you can get a sense of what information needs to be in it.

#3

for me, it would be terrific if I can perform classification using filterObject module so that I can overlay classified objects on the original image. So I wanted to export the rules in CPA to be used by filterObject… but you mean it’s impossible to do so depending on the type of classifiers, don’t you? sorry for my poor English…

and I think you are saying that it’s not possible to import objects. then after identifying classification rules using R or whatever, is it possible to write a rule file to be understood by filterObject?

#4

It depends on what classifier you tell CPA to use to group your cells. Right now you’re using RandomForest, which only produces a list of the important features, but not rules.:

If you change that dropdown option to FastGentleBoosting instead of RandomForest, then hit “Train” again, you’ll get a list of rules that you can copy into a text file, which can be read by FilterObjects.

Does that make more sense?

#5

thanks, that’s very kind of you. but what if I want other classifiers like random forest or SVM?

#6

If you want to run CP from source (or even the nightly build that’s on the website- it’s really out of date, but should be up-to-date enough for these purposes) you can export the classifier model from CPA and then load THAT into FilterObjects; I should add though that it’s experimental (I’ve personally never tried it) and may disappear from CP 3.0. That’s why I generally recommend using FGB instead, but if you’re willing to “live dangerously” you can give it a shot.

#7

thank you for your support always!