Finite-difference wave modeling with an object-oriented API (2nd attempt) #308
Conversation
This removes the compilation step and gives the same speed results.
Split into multiple files. Should work right now. Will populate with the new classes later.
leouieda
changed the title
|
A possible way of not relying on h5py is to optionally use memory-mapped numpy arrays for the cache. It seems that HDF5 has it's problems and Update: using numpy memmaps might be a problem because they are difficult to resize when running the simulation a second time. Would have to implement it and see what the performance is like. |
As given in the paper by Di Bartolo et al (2012).
Brought over the old classes and functions. Adapted the Equivalent Staggered Grid of Di Bartolo et al. (2012) that was implemented for the SH waves.
|
uow!!! I'm really going back to this ... esg acoustic, I think I can help on that I have done that for 3D here... |
|
@eusoubrasileiro yep! I'm reusing a lot of your functions from the previous PRs. I have separated things a bit more. The wavelets go into a Here is an example that I'm using for my geophysics class: http:https://mybinder.org/repo/leouieda/geofisica2/notebooks/notebooks/1-ondas-sismicas.ipynb I'm hopping to get this finished in October when I'll have some time off from work. |
|
interesting about the hdf5 problems... some of the suggestions on your link is to use numpy format *.npy I will try to take a deeper look in this soon. |
The progressbar uses the stdout stream, which can't be pickled. So can't use it with multiprocessing. Setting verbose=False should completely bypass the progressbar.
Calculate lambda and mu Lamé parameters from the P and S velocities and density. Functions transported from #308 with tests and added documentation.
Calculate lambda and mu Lamé parameters from the P and S velocities and density. Functions transported from #308 with tests and added documentation.
The classes set basic attributes of the wavelets and can be called (like
functions) to evaluate the wavelet at a given time. They also have a
`sample` method to sample with a given time interval.
For example:
wavelet = RickerWavelet(f=10)
value = wavelet(time=0.1)
values = wavelet(time=np.linspace(0, 1, 100))
samples = wavelet.sample(dt=0.001)
These classes can be used in the finite difference modeling as sources
and to generate samples of a wavelet for convolutional modeling.
Ports code over from #308
Defaults to mp4 and adds extra options for avconv.
It's available as a conda package.
Fixes #70
This is a retry at PR #137. I'll start fresh and take the best code and lessons learned from that PR to make this.
Defines a new API for the
fatiando.seismic.wavefdmodule. You now interact with simulation classes instead of generator functions. The simulation data that was thrown away before is now kept in an HDF5 file. This adds the dependency ofh5pybut it's worth it. The class also knows how to produce an animation of the simulation. The video animation can be embedded into the Jupyter notebook.As an added bonus, replaces the Cython time-stepping functions with numba equivalents without loss
in speed.
TODO
Checklist:
doc/install.rst,requirements.txt,environment.yml,ci/requirements-conda.txtandci/requirements-pip.txt.cd doc; makelocally)