def run(self, model_path: Path):
"""The model is downloaded when the run method is invoked.
Arguments:
model_path: The path to the trained model.
"""
# 1: Use the host and port at runtime so it works in the cloud.
# $ML_SERVER_URL/blob/master/mlserver/settings.py#L50
if self.version == 1:
# TODO: Reload the next version model of the model.
self.settings.update({"host": self.host, "http_port": self.port})
with open("settings.json", "w") as f:
json.dump(self.settings, f)
# 2. Store the model-settings
# $ML_SERVER_URL/blob/master/mlserver/settings.py#L120
self.model_settings["parameters"] = {
"version": f"v0.0.{self.version}",
"uri": str(model_path.absolute()),
}
with open("model-settings.json", "w") as f:
json.dump(self.model_settings, f)
# 3. Launch the Model Server
subprocess.Popen("mlserver start .", shell=True)
# 4. Increment the version for the next time run is called.
self.version += 1
else:
# TODO: Load the next model and unload the previous one.
pass
def run(self):
# 2: Load the Digits
digits = datasets.load_digits()
# 3: To apply a classifier on this data,
# we need to flatten the image, to
# turn the data in a (samples, feature) matrix:
n_samples = len(digits.images)
data = digits.images.reshape((n_samples, -1))
# 4: Create a classifier: a support vector classifier
classifier = svm.SVC(gamma=0.001)
# 5: Split data into train and test subsets
X_train, _, y_train, _ = train_test_split(data, digits.target, test_size=0.5, shuffle=False)
# 6: We learn the digits on the first half of the digits
classifier.fit(X_train, y_train)
# 7: Save the Sklearn model with `joblib`.
model_file_name = "mnist-svm.joblib"
joblib.dump(classifier, model_file_name)
# 8: Keep a reference the the generated model.
self.best_model_path = Path("mnist-svm.joblib")
def setup(
self,
trainer: "pl.Trainer",
pl_module: "pl.LightningModule",
stage: Optional[str] = None,
) -> None:
log_dir = self._get_logdir(trainer)
self.work.log_dir = Path(log_dir) if log_dir is not None else None
self._collect_logger_metadata(trainer)
def test_artifacts_tracker(tmpdir):
work = ScriptRunner(root_path=os.path.dirname(__file__), script_path=__file__)
tracker = PLAppArtifactsTracker(work=work)
trainer = Mock()
trainer.loggers = []
trainer.default_root_dir = "default_root_dir"
tracker.setup(trainer=trainer, pl_module=Mock())
assert work.log_dir == Path("default_root_dir")
assert not work.logger_metadatas
trainer.loggers = [TensorBoardLogger(save_dir=tmpdir)]
trainer.logger = trainer.loggers[0]
tracker.setup(trainer=trainer, pl_module=Mock())
assert work.log_dir == Path(tmpdir / "lightning_logs" / "version_0")
assert len(work.logger_metadatas) == 1
assert work.logger_metadatas[0] == {"class_name": "TensorBoardLogger"}
# call setup a second time and the metadata length should not change
tracker.setup(trainer=trainer, pl_module=Mock())
assert len(work.logger_metadatas) == 1
def _get_build_config(root_path: str) -> Optional[BuildConfig]:
# These are the requirements for the script runner itself
requirements = [
"protobuf<4.21.0",
"pytorch-lightning<=1.6.3",
"pydantic<=1.9.0",
]
if Path(root_path, "requirements.txt").exists():
# Requirements from the user's code folder
requirements.extend(
load_requirements(root_path, file_name="requirements.txt"))
return BuildConfig(requirements=requirements)
def run(self) -> None:
if not self.triggered:
return
if self.script_runner is None:
self.script_runner = ScriptRunner(
root_path=str(Path(__file__).parent / "source"),
script_path=str(
Path(__file__).parent / "source" / self.script_path),
script_args=self.script_args,
env=self._prepare_environment(),
parallel=True,
cloud_compute=CloudCompute(**self.cloud_compute_args),
raise_exception=False,
)
self.script_runner.run()
self.running = self.script_runner is not None and self.script_runner.has_started
self.succeeded = self.script_runner is not None and self.script_runner.has_succeeded
self.failed = self.script_runner is not None and self.script_runner.has_failed
if self.succeeded or self.failed:
self.triggered = False
def on_train_start(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule") -> None:
if trainer.checkpoint_callback and trainer.checkpoint_callback.dirpath is not None:
self.work.checkpoint_dir = Path(trainer.checkpoint_callback.dirpath)
def configure_layout(self):
return StaticWebFrontend(str(Path(__file__).parent / "ui/build"))