This project uses Python 3, and Tensorflow 2.
The first step to set it up is to install the required Python
libraries. These commands and the all the ones that follow are to be
run from the deepquake/ directory:
pip install --upgrade pip
pip install -r requirements.txt
Some configuration variables also need to be filled. Rename
deepquake/local_config.py.sample to
deepquake/local_config.py, and edit it, especially ROOT_PATH,
with the full path of the deepquake/ directory, and the MongoDB
credentials (for storing prediction results).
The input event and station information can be retrieved from FDSN webservices. For example, how to get input data from 2014 to 2018:
wget "http://rdsa.knmi.nl/fdsnws/event/1/query?starttime=2014-01-01&endtime=2019-01-01" -O inputdata/event_catalog.xml
wget "http://rdsa.knmi.nl/fdsnws/station/1/query?level=response&starttime=2014-01-01&endtime=2019-01-01" -O inputdata/NL_stations_2014-2018.xml
To fetch sample data use the deepquake/preprocess/get_streams.py
script. It gets streams either from FDSN webservices or local files,
processes them, and saves them in .tfrecords files. Before running
it, it is necessary to edit the SDS_ROOT variable in the
deepquake/preprocess/stream.py file with the path (to be) used
for locally storing MSEED files in a SDS-like directory structure.
To split the samples into train/test/evaluation sets, it is possible
to use the scripts/split_sets.py. Note that this might not the ideal
way to do it, because this script can put in separate sets samples
from different stations but corresponding to the same event.
Example:
python deepquake/preprocess/get_streams.py --inventory inputdata/NL_stations_2014-2018.xml --events inputdata/event_catalog.xml --output streams/ --event_fraction <0.01~1.0> --use-fdsn
python scripts/split_sets.py -i streams/filter/0 -o dataset/noise/
python scripts/split_sets.py -i streams/filter/1 -o dataset/events/
python scripts/split_sets.py -i streams/filter/2 -o dataset/events/
Example:
python deepquake/train.py --feature detection --num_epochs <N>
Example:
python deepquake/evaluate.py --feature detection <saved_weights_file.h5>
TODO: arguments
Example:
python deepquake/predict.py