Open blainerothrock opened 4 years ago
Running the initial testing script from the README resulted in bad results:
This is a test of SAC_20021102_XF
earthquake
Predict 0 1
Actual
0 35 0
1 42 15
unsure why.
Vivian shared a directory located on the GPGPU Machine which had a different version of that. This allowed for much better results that are closer the results reported in the paper.
{
"augmentations": "amplitude:noise",
"batch_size": 64,
"dataset_directory": "/home/b/dev/seis/vivian_earthquake/earthquakes/data/prepare_backup/trigger",
"embedding_size": 10,
"learning_rate": 2e-05,
"length": 8192,
"loss_function": "dpcl",
"model_type": "conv",
"num_epochs": 100,
"num_workers": 10,
"output_directory": "models/run12",
"sample_strategy": "sequential",
"split": "SAC_20021102_XF_prem",
"training_strategy": "siamese",
"transforms": "bandpass:whiten",
"weight_decay": 0.1
}
Predict 0 1
Actual
0 51 1
1 16 59
Using the current code base, reproduce the results of the paper
Data and installation instructions found in #1