wusunlab / chflux Goto Github PK

Timelag optimization is not robust. The following improvements are planned:

• Use the exponential function form for timelag search [WON'T FIX]
  • exp function did not work better than linearly transformed function; keep it but drop the intercept term [20 Apr 2017]
• Perform timelag optimization only when the concentration difference is large enough [WON'T FIX]
  • the decision of whether to trust the optimized timelag values shall be left to the user, not the developer [23 Apr 2017]
• Allow prescribed timelag values from external files
  • solved in the commit 2dd9d1e in the develop branch.

set_value() needs to be replaced with .at[] due to deprecation, although value assignment with .at[] is slower.

TODO: sort the list generated by glob.glob()

Calculate quality flag from the flux values derived from three fitting methods using Dixon's Q-test.

The function theilslopes() is two to three times slower than nonlinear regression. This is because the algorithm seems to have a time complexity of O(n²). Consider replacing Theil-Sen estimator with a more efficient robust linear regression method. (Perhaps an MM-estimator?)

Will restructure the main program in a modularized and object-oriented fashion