Closed tranqkhue closed 2 years ago
This problem still resists even trying:
mkl=2019.* blas=*=*mkl
blas=*=mkl
blas=*=openblas
And even trying MKL gives more error as it could not process nan values, as it diverges the polyfit function if nan values existThe problem is fixed by limiting the number of threads OpenBLAS export OPENBLAS_NUM_THREADS=1
Ref: https://github.com/numpy/numpy/issues/19057#issuecomment-845835734
Adding
scipy.signal.savgol_filter
consumes a lot of CPU usage The problem trace to https://github.com/scipy/scipy/blob/4cf21e753cf937d1c6c2d2a0e372fbc1dbbeea81/scipy/signal/_savitzky_golay.py#L193However, if we modify https://github.com/tranqkhue/damn_3dlidar/blob/f97604663e67618e0ecbf9a7fb85227c03bdb6ea/test2.py#L48 from
axis=1
toaxis=0
, the issue is gone. This is weird as number of calculations, as divided by a fixed-length window, are the same