ewellenr/earthquake_prediction_exploration
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Using public accessible data, retrievable through seismicportal.eu, create a forecaster for earthquake events (time, location, magnitude)
11 Commits 2 Branches 0 Tags 385 KiB
initial_exploration
Go to file
Ethan Wellenreiter 28096a932a Adding gitignore so that the acquired data isn't put into the repo and also the model logs aren't either 
If it's important enough, it can be placed in a different folder

Signed-off-by: Ethan Wellenreiter <ewellenreiter@gmail.com>
2025-06-23 13:29:17 -04:00
data_acquisition Moving data related stuff to a folder 2025-06-23 13:22:09 -04:00
.gitignore Adding gitignore so that the acquired data isn't put into the repo and also the model logs aren't either 2025-06-23 13:29:17 -04:00
full_regression_model.ipynb Current generation of model running 2025-06-23 13:24:04 -04:00
model_training.ipynb The model investigative jupyter notebook. It's older 2025-06-23 13:28:03 -04:00
README.md Adding/updating README 2025-06-23 13:27:38 -04:00
rnn_regression_model.py Current generation of model running 2025-06-23 13:24:04 -04:00

README.md

A repository for exploration into earthquake forecasting

Things to do

  1. Update the metric for the RNN to include an accuracy for high magnitude stuff
  2. Try farther look ahead models and/or just an amount of time ahead
    • can consider something like an auto-transformer type build where it feeds it's prediction back into itself until it's next prediction is outside of a time?

Frame the issue

Need to care more about the rare events. Let's say above mag 8. or mag 6.

Run command

(current best performance) C:/Users/ewell/anaconda3/envs/test_space/python.exe d:/projects/earthquake_prediction_exploration/rnn_regression_model.py --max_epochs 50 -x time_to_next_event lat lon depth mag --hidden_size 1000 --seq_size 100 --batch_size 1000