A student of mine has been working with FSL because we need to study our methodologies' effects on quantification results. We are fitting synthetic spectra. For each model we need to fit hundreds of spectra to get useful statistics from the analyses. Currently, 200 spectra is taking 5-7 hours, so 500-1000 will take forever for each model. While the actual fitting results may not match the simulation parameters perfectly, they are a good starting point and should cut down on the fitting time significantly. This is an especially relevant problem for deep learning research where we mainly use synthetic data.
Is this something that could be proto-typed rather quickly, at least in a forked repo? Do you have any suggestions for how to go about this? I am happy to contribute to a final, polished solution, but for the time being, a speedy solution would be very helpful.
Hello,
A student of mine has been working with FSL because we need to study our methodologies' effects on quantification results. We are fitting synthetic spectra. For each model we need to fit hundreds of spectra to get useful statistics from the analyses. Currently, 200 spectra is taking 5-7 hours, so 500-1000 will take forever for each model. While the actual fitting results may not match the simulation parameters perfectly, they are a good starting point and should cut down on the fitting time significantly. This is an especially relevant problem for deep learning research where we mainly use synthetic data.
Is this something that could be proto-typed rather quickly, at least in a forked repo? Do you have any suggestions for how to go about this? I am happy to contribute to a final, polished solution, but for the time being, a speedy solution would be very helpful.
Best, John