Hi, we are not allowed to redistribute the downloaded kik-net data, please directly download the kik-net data from their website kik-net and follow the preprocessing pipeline described in Bahrampouri. For the first step, you don't need to work on the kik-net dataset and can get started with your own dataset

In order to use our default training file: (1) dataset

1, the sampling frequency of the dataset should be fixed, (all records in 100Hz, or 200Hz, or 50Hz, etc)

2, the shape of your dataset should be : [N, 3, time*sampling_frequency], where N is the number of records

3, each waveform should start with the onset of P wave arrival

4, preprocessing is required, which is described in Bahrampouri and baseline correction of the dataset is required (available in `Python_libs/pylib_gm_processing.py`)

(2) metafile in csv format

(1) create a meta file associated with the dataset like:

Other notices: we train the model using velocity dataset and differentiate the synthetic velocity waveforms to get the accelerograms, so be sure use the baseline correction method to process the training dataset and apply the same baseline to your synthetic accelerograms. (built-in in the tutorial.ipynb). To quick start with training your model, I would suggest reduce the number of parameters . e.g just conditioned on magnitude or even unconditioned, and use a smaller dataset (10k~20k records), which will make it easier to prepare the dataset.

yzshi5 / GM-GANO

Dataset #1

1, the sampling frequency of the dataset should be fixed, (all records in 100Hz, or 200Hz, or 50Hz, etc)

2, the shape of your dataset should be : [N, 3, time*sampling_frequency], where N is the number of records

3, each waveform should start with the onset of P wave arrival

4, preprocessing is required, which is described in Bahrampouri and baseline correction of the dataset is required (available in `Python_libs/pylib_gm_processing.py`)

(1) create a meta file associated with the dataset like:

yzshi5 / GM-GANO

Dataset #1

1, the sampling frequency of the dataset should be fixed, (all records in 100Hz, or 200Hz, or 50Hz, etc)

2, the shape of your dataset should be : [N, 3, time*sampling_frequency], where N is the number of records

3, each waveform should start with the onset of P wave arrival

4, preprocessing is required, which is described in Bahrampouri and baseline correction of the dataset is required (available in Python_libs/pylib_gm_processing.py)

(1) create a meta file associated with the dataset like:

4, preprocessing is required, which is described in Bahrampouri and baseline correction of the dataset is required (available in `Python_libs/pylib_gm_processing.py`)