Large input data condition number problem #442
-
The objective So what should I do exactly? How should I pre-process it. and what does it mean by passing in augmented data? My data comes from the surface pressure of a moving body in CFD. Already tried tests I have a total of 10,000 snapshots. I tried to increase and decrease the amount of data as input but I still get the error msg. |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments
-
Hi @zonexo, I suggest trying to rescale the data by subtracting the temporal mean or the initial condition, for example. This operation will change the spectrum of the snapshots matrix and hopefully reduce the condition number. |
Beta Was this translation helpful? Give feedback.
-
Hello, Line 175 in 261dba9 , Lines 145 to 148 in 261dba9 ) i.e use a small value like 1e-6 (the value is problem-dependent) to regularize the minimization problem and avoid spurious modes due to ill-conditioning. See this tutorial |
Beta Was this translation helpful? Give feedback.
-
Hi all, thanks for the suggestions. I forgot to mention that I have tried the tikhonov_regularization, using 1e-8 as suggested in the tutorial. I will experiment with more values since I still get the error. I will also try to rescale the data. Thanks! |
Beta Was this translation helpful? Give feedback.
Hello,
You can also use the tikhonov parameter, (See
PyDMD/pydmd/dmdbase.py
Line 175 in 261dba9
,
PyDMD/pydmd/dmdbase.py
Lines 145 to 148 in 261dba9
)
i.e use a small value like 1e-6 (the value is problem-dependent) to regularize the minimization problem and avoid spurious modes due to ill-conditioning.
See this tutorial