Batch Prediction¶

1. Download demo data¶

cd PhaseNet
wget https://github.com/wayneweiqiang/PhaseNet/releases/download/test_data/test_data.zip
unzip test_data.zip

2. Run batch prediction¶

PhaseNet currently supports four data formats: mseed, sac, hdf5, and numpy.

• For mseed format:

  python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/mseed.csv --data_dir=test_data/mseed --format=mseed --plot_figure

• For sac format:

  python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/sac.csv --data_dir=test_data/sac --format=sac --plot_figure

• For numpy format:

  python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/npz.csv --data_dir=test_data/npz --format=numpy --plot_figure

• For hdf5 format:

  python phasenet/predict.py --model=model/190703-214543 --hdf5_file=test_data/data.h5 --hdf5_group=data --format=hdf5 --plot_figure

• For a seismic array (used by QuakeFlow):

  python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/mseed_array.csv --data_dir=test_data/mseed_array --stations=test_data/stations.json  --format=mseed_array --amplitude

  python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/mseed2.csv --data_dir=test_data/mseed --stations=test_data/stations.json  --format=mseed_array --amplitude

Notes:

1. Remove the "--plot_figure" argument for large datasets, because plotting can be very slow.

Optional arguments:

usage: predict.py [-h] [--batch_size BATCH_SIZE] [--model_dir MODEL_DIR]
                  [--data_dir DATA_DIR] [--data_list DATA_LIST]
                  [--hdf5_file HDF5_FILE] [--hdf5_group HDF5_GROUP]
                  [--result_dir RESULT_DIR] [--result_fname RESULT_FNAME]
                  [--min_p_prob MIN_P_PROB] [--min_s_prob MIN_S_PROB]
                  [--mpd MPD] [--amplitude] [--format FORMAT]
                  [--s3_url S3_URL] [--stations STATIONS] [--plot_figure]
                  [--save_prob]

optional arguments:
  -h, --help            show this help message and exit
  --batch_size BATCH_SIZE
                        batch size
  --model_dir MODEL_DIR
                        Checkpoint directory (default: None)
  --data_dir DATA_DIR   Input file directory
  --data_list DATA_LIST
                        Input csv file
  --hdf5_file HDF5_FILE
                        Input hdf5 file
  --hdf5_group HDF5_GROUP
                        data group name in hdf5 file
  --result_dir RESULT_DIR
                        Output directory
  --result_fname RESULT_FNAME
                        Output file
  --min_p_prob MIN_P_PROB
                        Probability threshold for P pick
  --min_s_prob MIN_S_PROB
                        Probability threshold for S pick
  --mpd MPD             Minimum peak distance
  --amplitude           if return amplitude value
  --format FORMAT       input format
  --stations STATIONS   seismic station info
  --plot_figure         If plot figure for test
  --save_prob           If save result for test

3. Output picks¶

• The output picks are saved to "results/picks.csv" on default

file_name	begin_time	station_id	phase_index	phase_time	phase_score	phase_amp	phase_type
2020-10-01T00:00*	2020-10-01T00:00:00.003	CI.BOM..HH	14734	2020-10-01T00:02:27.343	0.708	2.4998866231208325e-14	P
2020-10-01T00:00*	2020-10-01T00:00:00.003	CI.BOM..HH	15487	2020-10-01T00:02:34.873	0.416	2.4998866231208325e-14	S
2020-10-01T00:00*	2020-10-01T00:00:00.003	CI.COA..HH	319	2020-10-01T00:00:03.193	0.762	3.708662269972206e-14	P

Notes:

1. The phase_index means which data point is the pick in the original sequence. So phase_time = begin_time + phase_index / sampling rate. The default sampling_rate is 100Hz

3. Read P/S picks¶

PhaseNet currently outputs two format: CSV and JSON

In [1]:

import pandas as pd
import json
import os
PROJECT_ROOT = os.path.realpath(os.path.join(os.path.abspath(''), ".."))

import pandas as pd import json import os PROJECT_ROOT = os.path.realpath(os.path.join(os.path.abspath(''), ".."))

In [2]:

picks_csv = pd.read_csv(os.path.join(PROJECT_ROOT, "results/picks.csv"), sep="\t")
picks_csv.loc[:, 'p_idx'] = picks_csv["p_idx"].apply(lambda x: x.strip("[]").split(","))
picks_csv.loc[:, 'p_prob'] = picks_csv["p_prob"].apply(lambda x: x.strip("[]").split(","))
picks_csv.loc[:, 's_idx'] = picks_csv["s_idx"].apply(lambda x: x.strip("[]").split(","))
picks_csv.loc[:, 's_prob'] = picks_csv["s_prob"].apply(lambda x: x.strip("[]").split(","))
print(picks_csv.iloc[1])
print(picks_csv.iloc[0])

picks_csv = pd.read_csv(os.path.join(PROJECT_ROOT, "results/picks.csv"), sep="\t") picks_csv.loc[:, 'p_idx'] = picks_csv["p_idx"].apply(lambda x: x.strip("[]").split(",")) picks_csv.loc[:, 'p_prob'] = picks_csv["p_prob"].apply(lambda x: x.strip("[]").split(",")) picks_csv.loc[:, 's_idx'] = picks_csv["s_idx"].apply(lambda x: x.strip("[]").split(",")) picks_csv.loc[:, 's_prob'] = picks_csv["s_prob"].apply(lambda x: x.strip("[]").split(",")) print(picks_csv.iloc[1]) print(picks_csv.iloc[0])

fname      NC.MCV..EH.0361339.npz
t0        1970-01-01T00:00:00.000
p_idx                [5999, 9015]
p_prob             [0.987, 0.981]
s_idx                [6181, 9205]
s_prob             [0.553, 0.873]
Name: 1, dtype: object
fname      NN.LHV..EH.0384064.npz
t0        1970-01-01T00:00:00.000
p_idx                          []
p_prob                         []
s_idx                          []
s_prob                         []
Name: 0, dtype: object

In [3]:

with open(os.path.join(PROJECT_ROOT, "results/picks.json")) as fp:
    picks_json = json.load(fp)  
print(picks_json[1])
print(picks_json[0])

with open(os.path.join(PROJECT_ROOT, "results/picks.json")) as fp: picks_json = json.load(fp) print(picks_json[1]) print(picks_json[0])

{'id': 'NC.MCV..EH.0361339.npz', 'timestamp': '1970-01-01T00:01:30.150', 'prob': 0.9811667799949646, 'type': 'p'}
{'id': 'NC.MCV..EH.0361339.npz', 'timestamp': '1970-01-01T00:00:59.990', 'prob': 0.9872905611991882, 'type': 'p'}