`TFCGAN` is a python package for strong ground motion simulation in time-frequency domain (TF) based on based on conditional generative adversarial network and phase retrieval approaches. Used as command line application, TFCGAN can be used to create .npy files of synthetic seismic waveforms using a pre-trained ANN model. ### Install Install python 3.10.0 (anaconda or venv), then: Standard install: ```commandline pip install --upgrade pip setuptools && pip install . ``` Developers install (install editable package and tests libraries): ```commandline pip install --upgrade pip setuptools && pip install -e ".[dev]" ``` In the Python virtualenv and then: Check versions: `pip freeze` should give `tensorflow==2.11.1` and `numpy==1.26.1` For developers, run tests via `pytest ./tests` ### Usage TFCGAN is a command line application to be launched from the terminal. For help, type `tfcgan --help` in your terminal TFCGAN can also be used in your Python code (disclaimer: the snippet below has not been tested, please provide feedback in case of import errors): ```python from tfcgan.tfcgan import TFCGAN # setup your parameters. Example: mag = 7 dist = 100 vs30 = 760 num_waveforms = 10 # Model tfcgan = TFCGAN() # Generate labels for a specific scenario tfcgan.create_scenario(mag, dist, vs30, num_waveforms = num_waveforms) # Generate the time-frequency representation tf = model.get_tf_representation # Generate waveform data t, data = tfcgan.get_ground_shaking_synthesis # Generate Fourier amplitude spectra freq, fas = tfcgan.get_fas_response # data is a Numpy Matrix of shape (num_waveforms, 4000). # Each waveform delta time is 0.01 sec # (i.e., each waveform is 40s long by default) ``` ### TODO list: - create model with tensorflow 12.16 or whatever, - create models in hdf format (see tensorflow doc) - check the running code by running the tests Performance ----------- Citation ----------

Esfahani, Reza DD, Fabrice Cotton, Matthias Ohrnberger, and Frank Scherbaum. 
"TFCGAN: Nonstationary Ground‐Motion Simulation in the Time–Frequency Domain 
Using Conditional Generative Adversarial Network (CGAN) and Phase Retrieval Methods." 
*Bulletin of the Seismological Society of America * 113, no. 1 (2023): 453-467. 
https://doi.org/10.1785/0120220068