Initial import
This commit is contained in:
commit
1ff9b17202
3
.gitignore
vendored
Normal file
3
.gitignore
vendored
Normal file
@ -0,0 +1,3 @@
|
||||
.python-version
|
||||
dist
|
||||
*.pyc
|
165
README.md
Normal file
165
README.md
Normal file
@ -0,0 +1,165 @@
|
||||
# P and S Waves Detection with Deep Learning
|
||||
|
||||
## Installation
|
||||
|
||||
```
|
||||
pip install --extra-index-url https://epos-apps.grid.cyfronet.pl/api/packages/epos-ai/pypi/simple epos_ai_picking_tools
|
||||
```
|
||||
|
||||
## Generalized Phase Detection (GPD) Model
|
||||
|
||||
### Documentation
|
||||
|
||||
<!-- [[[cog
|
||||
import cog
|
||||
from epos_ai_picking_tools import cli_gpd as cli
|
||||
from click.testing import CliRunner
|
||||
runner = CliRunner()
|
||||
result = runner.invoke(cli.cli, ["--help"])
|
||||
help = result.output.replace("Usage: cli", "Usage: gpd_tool")
|
||||
cog.out(
|
||||
"```\n{}\n```".format(help)
|
||||
)
|
||||
]]] -->
|
||||
```
|
||||
Usage: gpd_tool [OPTIONS] COMMAND [ARGS]...
|
||||
|
||||
Options:
|
||||
--help Show this message and exit.
|
||||
|
||||
Commands:
|
||||
citation Prints citation of the model
|
||||
list-pretrained Show pretrained model names
|
||||
pick Detect phases in streams
|
||||
version Prints version
|
||||
|
||||
```
|
||||
<!-- [[[end]]] -->
|
||||
|
||||
### Pick
|
||||
|
||||
<!-- [[[cog
|
||||
import cog
|
||||
from epos_ai_picking_tools import cli_gpd as cli
|
||||
from click.testing import CliRunner
|
||||
runner = CliRunner()
|
||||
result = runner.invoke(cli.cli, ["pick", "--help"])
|
||||
help = result.output.replace("Usage: cli", "Usage: gpd_tool")
|
||||
cog.out(
|
||||
"```\n{}\n```".format(help)
|
||||
)
|
||||
]]] -->
|
||||
```
|
||||
Usage: gpd_tool pick [OPTIONS] [STREAM_FILE_NAMES]...
|
||||
|
||||
Detect phases in streams
|
||||
|
||||
Options:
|
||||
-w, --weights TEXT for possible options see output of 'list-pretrained'
|
||||
[default: original]
|
||||
-o, --output PATH directory to store results [default: .]
|
||||
-s, --stride INTEGER stride in samples for point prediction models
|
||||
[default: 10]
|
||||
-tp, --threshold-p FLOAT detection threshold for the P phase [default: 0.75]
|
||||
-ts, --threshold-s FLOAT detection threshold for the S phase [default: 0.75]
|
||||
--help Show this message and exit.
|
||||
|
||||
```
|
||||
<!-- [[[end]]] -->
|
||||
|
||||
### Citation
|
||||
|
||||
<!-- [[[cog
|
||||
import cog
|
||||
from epos_ai_picking_tools import cli_gpd as cli
|
||||
from click.testing import CliRunner
|
||||
runner = CliRunner()
|
||||
result = runner.invoke(cli.cli, ["citation"])
|
||||
help = result.output.replace("Usage: cli", "Usage: gpd_tool")
|
||||
cog.out(
|
||||
"\n{}\n".format(help)
|
||||
)
|
||||
]]] -->
|
||||
|
||||
Ross, Z. E., Meier, M.-A., Hauksson, E., & Heaton, T. H. (2018). Generalized Seismic Phase Detection with Deep Learning. ArXiv:1805.01075 [Physics]. https://arxiv.org/abs/1805.01075
|
||||
|
||||
<!-- [[[end]]] -->
|
||||
|
||||
## PhaseNet
|
||||
|
||||
### Documentation
|
||||
|
||||
<!-- [[[cog
|
||||
import cog
|
||||
from epos_ai_picking_tools import cli_phasenet as cli
|
||||
from click.testing import CliRunner
|
||||
runner = CliRunner()
|
||||
result = runner.invoke(cli.cli, ["--help"])
|
||||
help = result.output.replace("Usage: cli", "Usage: phasenet_tool")
|
||||
cog.out(
|
||||
"```\n{}\n```".format(help)
|
||||
)
|
||||
]]] -->
|
||||
```
|
||||
Usage: phasenet_tool [OPTIONS] COMMAND [ARGS]...
|
||||
|
||||
Options:
|
||||
--help Show this message and exit.
|
||||
|
||||
Commands:
|
||||
citation Prints citation of the model
|
||||
list-pretrained Show pretrained model names
|
||||
pick Detect phases in streams
|
||||
version Prints version
|
||||
|
||||
```
|
||||
<!-- [[[end]]] -->
|
||||
|
||||
### Pick
|
||||
|
||||
<!-- [[[cog
|
||||
import cog
|
||||
from epos_ai_picking_tools import cli_phasenet as cli
|
||||
from click.testing import CliRunner
|
||||
runner = CliRunner()
|
||||
result = runner.invoke(cli.cli, ["pick", "--help"])
|
||||
help = result.output.replace("Usage: cli", "Usage: phasenet_tool")
|
||||
cog.out(
|
||||
"```\n{}\n```".format(help)
|
||||
)
|
||||
]]] -->
|
||||
```
|
||||
Usage: phasenet_tool pick [OPTIONS] [STREAM_FILE_NAMES]...
|
||||
|
||||
Detect phases in streams
|
||||
|
||||
Options:
|
||||
-w, --weights TEXT for possible options see output of 'list-pretrained'
|
||||
[default: original]
|
||||
-o, --output PATH directory to store results [default: .]
|
||||
-tp, --threshold-p FLOAT detection threshold for the P phase [default: 0.3]
|
||||
-ts, --threshold-s FLOAT detection threshold for the S phase [default: 0.3]
|
||||
-b, --blinding TUPLE number of prediction samples to discard on each side
|
||||
of each window prediction [default: 0, 0]
|
||||
--help Show this message and exit.
|
||||
|
||||
```
|
||||
<!-- [[[end]]] -->
|
||||
|
||||
### Citation
|
||||
|
||||
<!-- [[[cog
|
||||
import cog
|
||||
from epos_ai_picking_tools import cli_phasenet as cli
|
||||
from click.testing import CliRunner
|
||||
runner = CliRunner()
|
||||
result = runner.invoke(cli.cli, ["citation"])
|
||||
help = result.output.replace("Usage: cli", "Usage: phasenet_tool")
|
||||
cog.out(
|
||||
"\n{}\n".format(help)
|
||||
)
|
||||
]]] -->
|
||||
|
||||
Zhu, W., & Beroza, G. C. (2019). PhaseNet: a deep-neural-network-based seismic arrival-time picking method. Geophysical Journal International, 216(1), 261-273. https://doi.org/10.1093/gji/ggy423
|
||||
|
||||
<!-- [[[end]]] -->
|
65
pyproject.toml
Normal file
65
pyproject.toml
Normal file
@ -0,0 +1,65 @@
|
||||
[build-system]
|
||||
requires = ["hatchling"]
|
||||
build-backend = "hatchling.build"
|
||||
|
||||
[project]
|
||||
name = "epos_ai_picking_tools"
|
||||
description = 'P and S Waves Detection with Deep Learning'
|
||||
readme = "README.md"
|
||||
requires-python = ">=3.7"
|
||||
license = "MIT"
|
||||
keywords = []
|
||||
authors = [
|
||||
{ name = "Hubert Siejkowski", email = "h.siejkowski@cyfronet.pl" },
|
||||
]
|
||||
classifiers = [
|
||||
"Programming Language :: Python",
|
||||
"Programming Language :: Python :: 3.7",
|
||||
"Programming Language :: Python :: 3.8",
|
||||
"Programming Language :: Python :: 3.9",
|
||||
"Programming Language :: Python :: 3.10",
|
||||
"Programming Language :: Python :: 3.11",
|
||||
"Programming Language :: Python :: Implementation :: CPython",
|
||||
]
|
||||
dependencies = [
|
||||
"seisbench==0.4.*",
|
||||
"click"
|
||||
]
|
||||
dynamic = ["version"]
|
||||
|
||||
[tool.hatch.version]
|
||||
path = "src/epos_ai_picking_tools/__about__.py"
|
||||
|
||||
[tool.hatch.envs.default]
|
||||
dependencies = [
|
||||
"pytest",
|
||||
"pytest-cov",
|
||||
"ipython",
|
||||
"black",
|
||||
"isort",
|
||||
"cogapp"
|
||||
]
|
||||
[tool.hatch.envs.default.scripts]
|
||||
cov = "pytest --cov-report=term-missing --cov-config=pyproject.toml --cov=gpd_tool --cov=tests"
|
||||
no-cov = "cov --no-cov"
|
||||
|
||||
[[tool.hatch.envs.test.matrix]]
|
||||
python = ["310", "311"]
|
||||
|
||||
[tool.coverage.run]
|
||||
branch = true
|
||||
parallel = true
|
||||
omit = [
|
||||
"gpd_tool/__about__.py",
|
||||
]
|
||||
|
||||
[tool.coverage.report]
|
||||
exclude_lines = [
|
||||
"no cov",
|
||||
"if __name__ == .__main__.:",
|
||||
"if TYPE_CHECKING:",
|
||||
]
|
||||
|
||||
[project.scripts]
|
||||
gpd_tool = "epos_ai_picking_tools.cli_gpd:cli"
|
||||
phasenet_tool = "epos_ai_picking_tools.cli_phasenet:cli"
|
@ -0,0 +1,29 @@
|
||||
import pathlib
|
||||
|
||||
import seisbench.models as sbm
|
||||
|
||||
from .model_runner import ModelRunner
|
||||
|
||||
|
||||
class GPDModelRunner(ModelRunner):
|
||||
|
||||
model_type = "GPD"
|
||||
|
||||
def __init__(self, weights_name="original", output_dir=pathlib.Path("."), **kwargs):
|
||||
self.model_name = getattr(sbm, GPDModelRunner.model_type)
|
||||
super(GPDModelRunner, self).__init__(
|
||||
weights_name=weights_name, output_dir=output_dir, **kwargs
|
||||
)
|
||||
|
||||
def process_kwargs(self, **kwargs):
|
||||
self.stride = int(kwargs.get("stride", 10))
|
||||
self.threshold_p = float(kwargs.get("threshold_p", 0.75))
|
||||
self.threshold_s = float(kwargs.get("threshold_s", 0.75))
|
||||
|
||||
self.annotate_kwargs = {
|
||||
"stride": self.stride,
|
||||
}
|
||||
|
||||
self.classify_kwargs = self.model.default_args.copy()
|
||||
self.classify_kwargs["P_threshold"] = self.threshold_p
|
||||
self.classify_kwargs["S_threshold"] = self.threshold_s
|
@ -0,0 +1,102 @@
|
||||
import json
|
||||
import pathlib
|
||||
import sys
|
||||
from collections import defaultdict
|
||||
|
||||
import obspy
|
||||
import seisbench.models as sbm
|
||||
from obspy.core.event import Catalog, Event, Pick, WaveformStreamID
|
||||
from obspy.io.json.default import Default
|
||||
|
||||
|
||||
def list_pretrained_models(model_runner_class):
|
||||
m = getattr(sbm, model_runner_class.model_type)
|
||||
weights = m.list_pretrained()
|
||||
return weights
|
||||
|
||||
|
||||
def exit_error(msg):
|
||||
print("ERROR:", msg)
|
||||
sys.exit(1)
|
||||
|
||||
|
||||
class ModelRunner:
|
||||
model_type = "EMPTY"
|
||||
|
||||
def __init__(self, weights_name="original", output_dir=pathlib.Path("."), **kwargs):
|
||||
# self.model_name = getattr(sbm, __class__.model_type)
|
||||
self.model = self.load_model(weights_name)
|
||||
self.output_dir = pathlib.Path(output_dir)
|
||||
self.process_kwargs(**kwargs)
|
||||
|
||||
def process_kwargs(self, **kwargs):
|
||||
pass
|
||||
|
||||
def citation(self):
|
||||
return self.model.citation
|
||||
|
||||
def load_model(self, weights_name):
|
||||
return self.model_name.from_pretrained(weights_name)
|
||||
|
||||
def load_stream(self, stream_file_name):
|
||||
return obspy.read(stream_file_name)
|
||||
|
||||
def save_picks(self, classs_picks, stream_path):
|
||||
dict_picks = list(map(lambda p: p.__dict__, classs_picks))
|
||||
fpath = self.output_dir / f"{stream_path.stem}_picks.json"
|
||||
|
||||
with open(fpath, "w") as fp:
|
||||
json.dump(dict_picks, fp, default=Default())
|
||||
|
||||
def save_quakeml(self, classs_picks, stream_path):
|
||||
e = Event()
|
||||
for cpick in classs_picks:
|
||||
net, sta, loc = cpick.trace_id.split(".")
|
||||
p = Pick(
|
||||
time=cpick.peak_time,
|
||||
phase_hint=cpick.phase,
|
||||
waveform_id=WaveformStreamID(
|
||||
network_code=net, station_code=sta, location_code=loc
|
||||
),
|
||||
)
|
||||
e.picks.append(p)
|
||||
|
||||
cat = Catalog([e])
|
||||
fpath = self.output_dir / f"{stream_path.stem}_picks.xml"
|
||||
cat.write(fpath, format="QUAKEML")
|
||||
|
||||
def write_annotations(self, annotations, stream_path):
|
||||
ann = annotations.copy()
|
||||
for tr in ann:
|
||||
tr.stats.channel = f"G_{tr.stats.component}"
|
||||
fpath = self.output_dir / f"{stream_path.stem}_annotations.mseed"
|
||||
ann.write(fpath, format="MSEED")
|
||||
|
||||
@staticmethod
|
||||
def validate_stream(stream):
|
||||
groups = defaultdict(list)
|
||||
for trace in stream:
|
||||
groups[trace.stats.station].append(trace.stats.channel[-1])
|
||||
|
||||
number_of_channels = list(map(len, groups.values()))
|
||||
|
||||
if max(number_of_channels) < 3:
|
||||
exit_error("Not enough traces in the stream")
|
||||
|
||||
def find_picks(self, stream_file_name, save_annotations=True):
|
||||
stream_path = pathlib.Path(stream_file_name)
|
||||
stream = self.load_stream(stream_path)
|
||||
|
||||
self.validate_stream(stream)
|
||||
|
||||
annotations = self.model.annotate(stream, **self.annotate_kwargs)
|
||||
|
||||
if save_annotations:
|
||||
self.write_annotations(annotations, stream_path)
|
||||
|
||||
classs_picks = self.model.classify_aggregate(annotations, self.classify_kwargs)
|
||||
|
||||
self.save_picks(classs_picks, stream_path)
|
||||
self.save_quakeml(classs_picks, stream_path)
|
||||
|
||||
return classs_picks
|
4
src/epos_ai_picking_tools/__about__.py
Normal file
4
src/epos_ai_picking_tools/__about__.py
Normal file
@ -0,0 +1,4 @@
|
||||
# SPDX-FileCopyrightText: 2022-present Hubert Siejkowski <h.siejkowski@gmail.com>
|
||||
#
|
||||
# SPDX-License-Identifier: MIT
|
||||
__version__ = "0.4.0"
|
0
src/epos_ai_picking_tools/__init__.py
Normal file
0
src/epos_ai_picking_tools/__init__.py
Normal file
90
src/epos_ai_picking_tools/cli_gpd.py
Normal file
90
src/epos_ai_picking_tools/cli_gpd.py
Normal file
@ -0,0 +1,90 @@
|
||||
import pathlib
|
||||
|
||||
import click
|
||||
import seisbench
|
||||
|
||||
from .__about__ import __version__
|
||||
from .gpd import GPDModelRunner
|
||||
from .model_runner import list_pretrained_models
|
||||
|
||||
|
||||
@click.group()
|
||||
def cli():
|
||||
pass
|
||||
|
||||
|
||||
@cli.command
|
||||
def version():
|
||||
"""Prints version"""
|
||||
print(f"SeisBench v{seisbench.__version__}")
|
||||
print(f"gpd_tool v{__version__}")
|
||||
|
||||
|
||||
@cli.command
|
||||
def citation():
|
||||
"""Prints citation of the model"""
|
||||
m = GPDModelRunner()
|
||||
print(m.citation())
|
||||
|
||||
|
||||
@cli.command
|
||||
def list_pretrained():
|
||||
"""Show pretrained model names"""
|
||||
print(", ".join(list_pretrained_models(GPDModelRunner)))
|
||||
|
||||
|
||||
@cli.command
|
||||
@click.option(
|
||||
"-w",
|
||||
"--weights",
|
||||
default="original",
|
||||
type=str,
|
||||
show_default=True,
|
||||
help=f"for possible options see output of 'list-pretrained'",
|
||||
)
|
||||
@click.option(
|
||||
"-o",
|
||||
"--output",
|
||||
default=pathlib.Path("."),
|
||||
type=click.Path(dir_okay=True, path_type=pathlib.Path),
|
||||
show_default=True,
|
||||
help="directory to store results",
|
||||
)
|
||||
@click.option(
|
||||
"-s",
|
||||
"--stride",
|
||||
default=10,
|
||||
type=int,
|
||||
show_default=True,
|
||||
help="stride in samples for point prediction models",
|
||||
)
|
||||
@click.option(
|
||||
"-tp",
|
||||
"--threshold-p",
|
||||
default=0.75,
|
||||
type=float,
|
||||
show_default=True,
|
||||
help="detection threshold for the P phase",
|
||||
)
|
||||
@click.option(
|
||||
"-ts",
|
||||
"--threshold-s",
|
||||
default=0.75,
|
||||
type=float,
|
||||
show_default=True,
|
||||
help="detection threshold for the S phase",
|
||||
)
|
||||
@click.argument("stream_file_names", nargs=-1, type=click.Path(exists=True))
|
||||
def pick(stream_file_names, weights, output, stride, threshold_p, threshold_s):
|
||||
"""Detect phases in streams"""
|
||||
if not output.exists():
|
||||
output.mkdir()
|
||||
m = GPDModelRunner(
|
||||
weights,
|
||||
output_dir=output,
|
||||
stride=stride,
|
||||
threshold_p=threshold_p,
|
||||
threshold_s=threshold_s,
|
||||
)
|
||||
for stream in stream_file_names:
|
||||
m.find_picks(stream)
|
90
src/epos_ai_picking_tools/cli_phasenet.py
Normal file
90
src/epos_ai_picking_tools/cli_phasenet.py
Normal file
@ -0,0 +1,90 @@
|
||||
import pathlib
|
||||
|
||||
import click
|
||||
import seisbench
|
||||
|
||||
from .__about__ import __version__
|
||||
from .model_runner import list_pretrained_models
|
||||
from .phasenet import PhaseNetModelRunner
|
||||
|
||||
|
||||
@click.group()
|
||||
def cli():
|
||||
pass
|
||||
|
||||
|
||||
@cli.command
|
||||
def version():
|
||||
"""Prints version"""
|
||||
print(f"SeisBench v{seisbench.__version__}")
|
||||
print(f"phasenet_tool v{__version__}")
|
||||
|
||||
|
||||
@cli.command
|
||||
def citation():
|
||||
"""Prints citation of the model"""
|
||||
m = PhaseNetModelRunner()
|
||||
print(m.citation())
|
||||
|
||||
|
||||
@cli.command
|
||||
def list_pretrained():
|
||||
"""Show pretrained model names"""
|
||||
print(", ".join(list_pretrained_models(PhaseNetModelRunner)))
|
||||
|
||||
|
||||
@cli.command
|
||||
@click.option(
|
||||
"-w",
|
||||
"--weights",
|
||||
default="original",
|
||||
type=str,
|
||||
show_default=True,
|
||||
help=f"for possible options see output of 'list-pretrained'",
|
||||
)
|
||||
@click.option(
|
||||
"-o",
|
||||
"--output",
|
||||
default=pathlib.Path("."),
|
||||
type=click.Path(dir_okay=True, path_type=pathlib.Path),
|
||||
show_default=True,
|
||||
help="directory to store results",
|
||||
)
|
||||
@click.option(
|
||||
"-tp",
|
||||
"--threshold-p",
|
||||
default=0.3,
|
||||
type=float,
|
||||
show_default=True,
|
||||
help="detection threshold for the P phase",
|
||||
)
|
||||
@click.option(
|
||||
"-ts",
|
||||
"--threshold-s",
|
||||
default=0.3,
|
||||
type=float,
|
||||
show_default=True,
|
||||
help="detection threshold for the S phase",
|
||||
)
|
||||
@click.option(
|
||||
"-b",
|
||||
"--blinding",
|
||||
default=(0, 0),
|
||||
type=tuple,
|
||||
show_default=True,
|
||||
help="number of prediction samples to discard on each side of each window prediction",
|
||||
)
|
||||
@click.argument("stream_file_names", nargs=-1, type=click.Path(exists=True))
|
||||
def pick(stream_file_names, weights, output, threshold_p, threshold_s, blinding):
|
||||
"""Detect phases in streams"""
|
||||
if not output.exists():
|
||||
output.mkdir()
|
||||
m = PhaseNetModelRunner(
|
||||
weights,
|
||||
output_dir=output,
|
||||
threshold_p=threshold_p,
|
||||
threshold_s=threshold_s,
|
||||
blinding=blinding,
|
||||
)
|
||||
for stream in stream_file_names:
|
||||
m.find_picks(stream)
|
29
src/epos_ai_picking_tools/gpd.py
Normal file
29
src/epos_ai_picking_tools/gpd.py
Normal file
@ -0,0 +1,29 @@
|
||||
import pathlib
|
||||
|
||||
import seisbench.models as sbm
|
||||
|
||||
from .model_runner import ModelRunner
|
||||
|
||||
|
||||
class GPDModelRunner(ModelRunner):
|
||||
|
||||
model_type = "GPD"
|
||||
|
||||
def __init__(self, weights_name="original", output_dir=pathlib.Path("."), **kwargs):
|
||||
self.model_name = getattr(sbm, GPDModelRunner.model_type)
|
||||
super(GPDModelRunner, self).__init__(
|
||||
weights_name=weights_name, output_dir=output_dir, **kwargs
|
||||
)
|
||||
|
||||
def process_kwargs(self, **kwargs):
|
||||
self.stride = int(kwargs.get("stride", 10))
|
||||
self.threshold_p = float(kwargs.get("threshold_p", 0.75))
|
||||
self.threshold_s = float(kwargs.get("threshold_s", 0.75))
|
||||
|
||||
self.annotate_kwargs = {
|
||||
"stride": self.stride,
|
||||
}
|
||||
|
||||
self.classify_kwargs = self.model.default_args.copy()
|
||||
self.classify_kwargs["P_threshold"] = self.threshold_p
|
||||
self.classify_kwargs["S_threshold"] = self.threshold_s
|
102
src/epos_ai_picking_tools/model_runner.py
Normal file
102
src/epos_ai_picking_tools/model_runner.py
Normal file
@ -0,0 +1,102 @@
|
||||
import json
|
||||
import pathlib
|
||||
import sys
|
||||
from collections import defaultdict
|
||||
|
||||
import obspy
|
||||
import seisbench.models as sbm
|
||||
from obspy.core.event import Catalog, Event, Pick, WaveformStreamID
|
||||
from obspy.io.json.default import Default
|
||||
|
||||
|
||||
def list_pretrained_models(model_runner_class):
|
||||
m = getattr(sbm, model_runner_class.model_type)
|
||||
weights = m.list_pretrained()
|
||||
return weights
|
||||
|
||||
|
||||
def exit_error(msg):
|
||||
print("ERROR:", msg)
|
||||
sys.exit(1)
|
||||
|
||||
|
||||
class ModelRunner:
|
||||
model_type = "EMPTY"
|
||||
|
||||
def __init__(self, weights_name="original", output_dir=pathlib.Path("."), **kwargs):
|
||||
# self.model_name = getattr(sbm, __class__.model_type)
|
||||
self.model = self.load_model(weights_name)
|
||||
self.output_dir = pathlib.Path(output_dir)
|
||||
self.process_kwargs(**kwargs)
|
||||
|
||||
def process_kwargs(self, **kwargs):
|
||||
pass
|
||||
|
||||
def citation(self):
|
||||
return self.model.citation
|
||||
|
||||
def load_model(self, weights_name):
|
||||
return self.model_name.from_pretrained(weights_name)
|
||||
|
||||
def load_stream(self, stream_file_name):
|
||||
return obspy.read(stream_file_name)
|
||||
|
||||
def save_picks(self, classs_picks, stream_path):
|
||||
dict_picks = list(map(lambda p: p.__dict__, classs_picks))
|
||||
fpath = self.output_dir / f"{stream_path.stem}_picks.json"
|
||||
|
||||
with open(fpath, "w") as fp:
|
||||
json.dump(dict_picks, fp, default=Default())
|
||||
|
||||
def save_quakeml(self, classs_picks, stream_path):
|
||||
e = Event()
|
||||
for cpick in classs_picks:
|
||||
net, sta, loc = cpick.trace_id.split(".")
|
||||
p = Pick(
|
||||
time=cpick.peak_time,
|
||||
phase_hint=cpick.phase,
|
||||
waveform_id=WaveformStreamID(
|
||||
network_code=net, station_code=sta, location_code=loc
|
||||
),
|
||||
)
|
||||
e.picks.append(p)
|
||||
|
||||
cat = Catalog([e])
|
||||
fpath = self.output_dir / f"{stream_path.stem}_picks.xml"
|
||||
cat.write(fpath, format="QUAKEML")
|
||||
|
||||
def write_annotations(self, annotations, stream_path):
|
||||
ann = annotations.copy()
|
||||
for tr in ann:
|
||||
tr.stats.channel = f"G_{tr.stats.component}"
|
||||
fpath = self.output_dir / f"{stream_path.stem}_annotations.mseed"
|
||||
ann.write(fpath, format="MSEED")
|
||||
|
||||
@staticmethod
|
||||
def validate_stream(stream):
|
||||
groups = defaultdict(list)
|
||||
for trace in stream:
|
||||
groups[trace.stats.station].append(trace.stats.channel[-1])
|
||||
|
||||
number_of_channels = list(map(len, groups.values()))
|
||||
|
||||
if max(number_of_channels) < 3:
|
||||
exit_error("Not enough traces in the stream")
|
||||
|
||||
def find_picks(self, stream_file_name, save_annotations=True):
|
||||
stream_path = pathlib.Path(stream_file_name)
|
||||
stream = self.load_stream(stream_path)
|
||||
|
||||
self.validate_stream(stream)
|
||||
|
||||
annotations = self.model.annotate(stream, **self.annotate_kwargs)
|
||||
|
||||
if save_annotations:
|
||||
self.write_annotations(annotations, stream_path)
|
||||
|
||||
classs_picks = self.model.classify_aggregate(annotations, self.classify_kwargs)
|
||||
|
||||
self.save_picks(classs_picks, stream_path)
|
||||
self.save_quakeml(classs_picks, stream_path)
|
||||
|
||||
return classs_picks
|
28
src/epos_ai_picking_tools/phasenet.py
Normal file
28
src/epos_ai_picking_tools/phasenet.py
Normal file
@ -0,0 +1,28 @@
|
||||
import pathlib
|
||||
|
||||
import seisbench.models as sbm
|
||||
|
||||
from .model_runner import ModelRunner
|
||||
|
||||
|
||||
class PhaseNetModelRunner(ModelRunner):
|
||||
|
||||
model_type = "PhaseNet"
|
||||
|
||||
def __init__(self, weights_name="original", output_dir=pathlib.Path("."), **kwargs):
|
||||
self.model_name = getattr(sbm, PhaseNetModelRunner.model_type)
|
||||
super(PhaseNetModelRunner, self).__init__(
|
||||
weights_name=weights_name, output_dir=output_dir, **kwargs
|
||||
)
|
||||
|
||||
def process_kwargs(self, **kwargs):
|
||||
self.threshold_p = float(kwargs.get("threshold_p", 0.3))
|
||||
self.threshold_s = float(kwargs.get("threshold_s", 0.3))
|
||||
self.blinding = kwargs.get("blinding", (0, 0))
|
||||
|
||||
self.annotate_kwargs = {}
|
||||
|
||||
self.classify_kwargs = self.model.default_args.copy()
|
||||
self.classify_kwargs["P_threshold"] = self.threshold_p
|
||||
self.classify_kwargs["S_threshold"] = self.threshold_s
|
||||
self.classify_kwargs["blinding"] = self.blinding
|
0
tests/__init__.py
Normal file
0
tests/__init__.py
Normal file
BIN
tests/dummy.mseed
Normal file
BIN
tests/dummy.mseed
Normal file
Binary file not shown.
85
tests/test_epos_ai_picking_tools.py
Normal file
85
tests/test_epos_ai_picking_tools.py
Normal file
@ -0,0 +1,85 @@
|
||||
import pathlib
|
||||
|
||||
import pytest
|
||||
import seisbench.models as sbm
|
||||
|
||||
from epos_ai_picking_tools.gpd import GPDModelRunner
|
||||
from epos_ai_picking_tools.phasenet import PhaseNetModelRunner
|
||||
|
||||
models = [
|
||||
(GPDModelRunner, "stead", sbm.GPD),
|
||||
(PhaseNetModelRunner, "original", sbm.PhaseNet),
|
||||
]
|
||||
|
||||
|
||||
@pytest.fixture()
|
||||
def simple_seed():
|
||||
current_dir = pathlib.Path(__file__).parent
|
||||
seed = current_dir / "dummy.mseed"
|
||||
return seed
|
||||
|
||||
|
||||
@pytest.fixture()
|
||||
def phasenet_model_runner(tmpdir):
|
||||
m = PhaseNetModelRunner("original", output_dir=tmpdir)
|
||||
return m
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model_runner,weights,seisbench_model", models)
|
||||
def test_fail_ModelRunner_load_weights(model_runner, weights, seisbench_model):
|
||||
with pytest.raises(ValueError):
|
||||
m = model_runner("no_exitsting_model_weights")
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model_runner,weights,seisbench_model", models)
|
||||
def test_ModelRunner_load_weights(model_runner, weights, seisbench_model):
|
||||
m = model_runner(weights)
|
||||
assert isinstance(m.model, seisbench_model)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model_runner,weights,seisbench_model", models)
|
||||
def test_ModelRunner_output_dir(model_runner, weights, seisbench_model):
|
||||
m = model_runner(weights)
|
||||
|
||||
assert m.output_dir == pathlib.Path(".")
|
||||
|
||||
m = model_runner(weights, "./tmp_dir")
|
||||
assert m.output_dir == pathlib.Path("./tmp_dir")
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model_runner,weights,seisbench_model", models)
|
||||
def test_ModelRunner_find_picks(
|
||||
model_runner, weights, seisbench_model, simple_seed, tmpdir
|
||||
):
|
||||
m = model_runner(weights, output_dir=tmpdir)
|
||||
|
||||
picks = m.find_picks(simple_seed)
|
||||
assert isinstance(picks, list)
|
||||
|
||||
json_in_dir = m.output_dir.glob("*.json")
|
||||
assert len(list(json_in_dir)) == 1
|
||||
|
||||
annotations_in_dir = m.output_dir.glob("*annotations*")
|
||||
assert len(list(annotations_in_dir)) == 1
|
||||
|
||||
|
||||
def test_GPDModelRunner_find_picks_reults(simple_seed, tmpdir):
|
||||
m = GPDModelRunner("stead", output_dir=tmpdir)
|
||||
picks = m.find_picks(simple_seed)
|
||||
|
||||
assert picks[0].phase == "P"
|
||||
assert picks[0].peak_time == "2000-01-01T07:00:05.100000Z"
|
||||
|
||||
assert picks[2].phase == "S"
|
||||
assert picks[2].peak_time == "2000-01-01T07:00:15.700000Z"
|
||||
|
||||
|
||||
def test_PhaseNetModelRunner_find_picks_reults(simple_seed, tmpdir):
|
||||
m = PhaseNetModelRunner("original", output_dir=tmpdir)
|
||||
picks = m.find_picks(simple_seed)
|
||||
|
||||
assert picks[0].phase == "P"
|
||||
assert picks[0].peak_time == "2000-01-01T07:00:05.350000Z"
|
||||
|
||||
assert picks[1].phase == "S"
|
||||
assert picks[1].peak_time == "2000-01-01T07:00:06.140000Z"
|
Loading…
Reference in New Issue
Block a user