Hi Paul, I worked on arrtifact reduction in FCS time trace data. There, I can predict different artifacts in time traces. Currently I am removing these artifacts by deleting the artifactual parts and contracting the trace, which works quite well, but introduces small "stitching" artifacts where you put the parts of the trace together. When discussing this approach at MAF2022, Thorsten Wohland and others brought up that it should be straight forward to instead assign the artifactual parts of a trace as np.nan and deal with it in the correlation algorithm by ignoring this value in the correlation and then adjusting the normalisation in the end. I have not dealt with the correlation algorithms that deeply yet - have you thought about the handling of NaN values e.g. in multipletau?
Hi Paul, I worked on arrtifact reduction in FCS time trace data. There, I can predict different artifacts in time traces. Currently I am removing these artifacts by deleting the artifactual parts and contracting the trace, which works quite well, but introduces small "stitching" artifacts where you put the parts of the trace together. When discussing this approach at MAF2022, Thorsten Wohland and others brought up that it should be straight forward to instead assign the artifactual parts of a trace as
np.nanand deal with it in the correlation algorithm by ignoring this value in the correlation and then adjusting the normalisation in the end. I have not dealt with the correlation algorithms that deeply yet - have you thought about the handling of NaN values e.g. inmultipletau?