Harmonic oscillation reduction

[sarajamal57]

To solve the overflow problem in issue #284 :

new py file cesium/features/lomb_scargle_norm.py incorporating in lomb_scargle_model()

• normalization of the input data if opt_normalize
• computation of the Lomb-Scargle model via fit_lomb_scargle()
• rescaling of the output dictionary via rescale_lomb_model()

Notes

1. modifications are solely made in cesium/features/lomb_scargle_norm.py not in the original source file cesium/features/lomb_scargle.py
2. successful local py tests for checking the rescaled output variables in the lomb_scargle_model.

Additional output
The full power spectrum and associated frequencies are added in the output dictionary in fit_lomb_scargle() [Lines 299-302]

[pep8speaks]

Hello @sarajamal57! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file cesium/features/lomb_scargle.py:

Line 56:81: E501 line too long (86 > 80 characters)
Line 59:5: E266 too many leading '#' for block comment
Line 164:5: E266 too many leading '#' for block comment
Line 173:81: E501 line too long (82 > 80 characters)
Line 384:5: E266 too many leading '#' for block comment
Line 386:81: E501 line too long (83 > 80 characters)
Line 390:5: E266 too many leading '#' for block comment
Line 401:81: E501 line too long (84 > 80 characters)
Line 402:81: E501 line too long (87 > 80 characters)
Line 407:81: E501 line too long (81 > 80 characters)
Line 438:9: W503 line break before binary operator
Line 439:9: W503 line break before binary operator
Line 476:9: W503 line break before binary operator
Line 485:81: E501 line too long (86 > 80 characters)
Line 494:81: E501 line too long (87 > 80 characters)

Comment last updated at 2021-05-27 22:37:18 UTC

[acrellin]

@sarajamal57 Reviving this long-neglected PR (sorry!). I've moved your changes into the existing lomb_scargle.py file -- does the diff here look right to you? And thanks again for your contribution!

[stefanv]

Extra parameter needs a docstring entry

[stefanv]

docstring needs formatting

[stefanv]

What does opt_normalize do?

We also need tests for these new flags.

[sarajamal57]

The normalization resolves the numerical overflow observed for instance in issue #284

Modified/Added functions

• lomb_scargle_model()
• rescale_lomb_model()

Procedure

If opt_normalize==False, this refers to the original Lomb-Scargle periodogram function.

Otherwise:
Step 1 - the time-serie (TS) is normalized, and the (mean, std scale) are stored
Step 2 - Lomb-Scargle periodogram & the best fitted model are computed on the normalized (signal, error)
Step 3 - Results are scaled back to the original/raw format using the stored (mean, scale)

Step 1
In function lomb_scargle_model():

### Normalization ###
    mmean  = np.nanmean(signal)
    mscale = np.nanstd(signal)
    if opt_normalize:
        signal = (signal-mmean)/mscale
        error  = error/mscale

Step 2
The lomb-scargle (LS) periodogram is estimated on the normalized time-serie (signal, associated error).

Step 3
The best fitted Lomb-Scargle model is rescaled back.
In function lomb_scargle_model() :

current_fit = rescale_lomb_model(fit, mmean, mscale) if (opt_normalize&(i==0)) else fit 
## function rescale_lomb_model() is used to rescale the best fitted model (computed on normalized TS)

The first component (i==0) of var freq_fits requires to re-inject the scale

if (opt_normalize&(i==0)):
            model_dict['freq_fits'][-1]['resid']*=mscale

[profjsb]

@sarajamal57 will you be able to add the tests/comments as per @stefanv's review?

[sarajamal57]

yes, will do, no problem !
Commented text seems to be in place in the only modified file lomb_scargle_model() (cesium/features/lomb_scargle.py)
I will add a py test def test_lomb_scargle_normalize() to cesium/features/tests/test_lomb_scargle_features.py

[profjsb]

Closed in favor of #313