search

CRAVA / crava

6 stars 13 forks source link

Scaling of wavelets #258

Closed hflesche closed 7 years ago

hflesche commented 7 years ago

@RagnarHauge and @hgolsen: I have run Crava with otherwise otherwise identical cases:

  1. One in which I import wavelets from RokDoc and set yes
  2. One in which I let Crava estimate the wavelet
  3. One in which I use the wavelets estimated by Crava in the previous setup and set yes

Runs number 1 and 3 fail and number 2 succeeds. The error message is identical in the two cases:

****************************************************************************************************
*****    Loading and processing data failed. CRAVA has to stop due to the following errors:    *****
****************************************************************************************************
Error when setting up wavelets: 

ERROR: Cannot find global wavelet scale. The maximum reflection
       coefficient of 50141.11  is outside valid range of [-1, +1].

ERROR: Cannot find global wavelet scale. The maximum reflection
       coefficient of 51558.45  is outside valid range of [-1, +1].

ERROR: Cannot find global wavelet scale. The maximum reflection
       coefficient of 54677.69  is outside valid range of [-1, +1].

ERROR: Cannot find global wavelet scale. The maximum reflection
       coefficient of 59915.95  is outside valid range of [-1, +1].

ERROR: Cannot find global wavelet scale. The maximum reflection
       coefficient of 68056.80  is outside valid range of [-1, +1].

ERROR: Cannot find global wavelet scale. The maximum reflection
       coefficient of 80526.41  is outside valid range of [-1, +1].

The amplitude values in the wavelets from RokDoc and Crava have a ratio of about 10:1. Still the reflection coefficient is given with identical values in the two runs. The estimated scaling factors are not shown in the log files.

hflesche commented 7 years ago

By closer examination of the case it seems that the error message was slightly misleading. We have examined the logs and removed parts with outliers. There was also a case of formatting the wavelets that are used. The header from RokDoc is not accepted by Crava, and a manual editing has to be done. We are now able to run through an "estimation" action without any scaling of the wavelets from RokDoc, and S/N values are acceptable.