Skip to content

Latest commit

 

History

History
98 lines (61 loc) · 4.17 KB

TODO.md

File metadata and controls

98 lines (61 loc) · 4.17 KB

Roadmap

General

  • Authors: how do we decide to mention the codes authors? As the code is going to evolve, it is going to be interesting to have specific authors per functions, with associated email adress.

  • Docstrings: according to book The Hitchhiker's Guide to Python: Best Practices for Development, we should develop the code according the PEP8 standards, which is what the Black soft is doing to your code if these standards are not met. Within the same book, sphinx is presented as the most popular documentation tool that allow to generate a documentation webpage from the codes' docstrings. Is is based on the reStructured text formatting used in the docstring. The reStructured text format is quite flexible, and several docstring standards exist. In order to be as simple as possible, most of the code so far uses the numpy docstring style. Should we agree on this?

  • Documentation: based on the above comments, the documentation is generated from the code docstring. After installing sphinx, we should also install the following sub-packages:

    • sphinx_bootstrap_theme for webpage style
    • sphinx_execute_code for example execution on build

    The main advantage of using sphinx is for allowing inter-sphinx mapping for the documentation (referencing a function based on the alread-existing documentation of other libraries like numpy, etc. that are also coded in sphinx).

  • Setup: for installing the package, we have to declare what is required. I made a non-exhaustive list in the freshly created setup.py (a copy from this setup template).

  • verbose mode ? logger, tqdm

  • top-level function from examples?

    • obspy example coherence
    • 1-y long script on PdF

  • everything is a numpy array (np.save/z, np.load)

Data management

  • Remove methods for strange formats (h5read, matread). We want the users to whether feed covariance with an obspy.Stream object stream or an covseisnet.ArrayStream object.

  • for calculating the covariance, if we get a obspy.Stream object, we have to turn it into a covseisnet.ArrayStream, involing a synchronize routine (method of the data.py module).

  • the time propertry returns only 1 time vector from 1 of the traces, knowing that they should be exactly on the same sampling and length. If not, this property should trigger an error/warning.

  • the homegenize will become synchronize for synchronizing streams on same sampling and length automatically (with many options).

  • stop using savitzky_golay for smoothing. Move to scipy.signal.smooth.

  • stop using robust for claculating the mad. Move to scipy.signal.mad.

  • remove "method" ``kwargs, when smooth=1, it is onebit spectral whitening.

  • removed plot method.

Covariance

  • move fft to covariance
  • remark: fft really depends on parameters
  • francis if you want to try differrnt fft up to you
  • spectral_width et entropy sont dans coherence

Todo

  • Cyril parle de covariance non-hermitienne. Il faut définir les propriétés d'une telle matrice (comment dealer avec les vp non hermitiennes, comment on calcule la coherence dans ce cas).
  • Spectral lines
  • rank peut être une liste de ranks.
  • inclure iso et formulate synthetic according to it

Correlation

Todo

  • Sigma en secondes

## Net is the new antenna

  • FDSN
  • Exemples: montrer les correlation sur Kamchatka triées avec net.epicentral_dist_from.

Beamforming (planewave)

Location ("beam" sphérique)

Calcul ttimes

cov, net = np.load('cov.npz')
# cov.select()
# net.select()
cov = cov.eigenvectors(rank=0)
corr = cn.correlation.calculate(cov)

model = np.load('model.npz')
# dictionary with 'lon', 'lat', 'z'

"ttimes.npz"
ttimes = np.load(tties.npz)
ttimes[station1]
ttimes[station2]


# Define grid
likelihood = covseisnet.GridSearch(model, net, ttimes='/path/to/ttimes.npz', build=False)
likelihood.calculate(corr)
  • On prévoit de coder la sélection + tard (Francis!)