I have a properties files from CellProfiler containing a few images that have the most examples of the variety of cells I need to sort using classifier, and I created a training set from this properties file in CellProfiler analyst. However it will only let me load this training set when I use this properties file, which would mean that I’d have to regenerate the training set every time I get new images and they may not contain the variety of cells needed to generate accurate training models. Is there a way to get around this?
Training set objects have to be in the set you’re analyzing; there is no way around that, unfortunately. This leaves you with basically 3 main options:
- have a set of “for training” images you run every time alongside the new images in your new set; however, in order to directly load the SAME training set, you’ll need to make sure the image numbers don’t change, so you’ll need to ensure that (likely by making those images alphabetically first before any new data that. you run). Even if you couldn’t load the exact training set, you could still more-easily-than-fully-from-scratch retrain, if you know which images to pull which are your “training control images” (you could specify them with Metadata, for example).
- Append new images to the old database; there isn’t a way to do this DIRECTLY in CellProfiler, but if you’re mildly handy with doing database interaction in any of SQL, Python, R, etc it could probably be done without too much difficulty.
- Try to train a classifier with data from enough batches such that it’s robust enough to not need to be trained every time, and just applied with the FilterObjects tool.
Thank you so much Beth- I think #1 makes the most sense for me as I can change the names of these files to ensure that they’ll come alphabetically first.