luxonis_ml.tracker.LuxonisTracker

class documentation

class LuxonisTracker:

Constructor: LuxonisTracker(project_name, project_id, run_name, run_id, ...)

Logger wrapper for TensorBoard, WandB, and MLflow.

LuxonisTracker stores run metadata, initializes the selected logging integrations lazily, and keeps a local fallback cache for MLflow logs that fail transiently.

Static Method	`rank_zero_only`	Wrap a function so only processes with rank=0 execute it.
Method	`__init__`	Create a tracker for one or more logging integrations.
Method	`close`	Finalize logging and save unsent logs locally.
Method	`log_hyperparams`	Log a hyperparameter dictionary.
Method	`log_image`	Log one image.
Method	`log_images`	Log multiple images.
Method	`log_matrix`	Log a matrix to the enabled logging services.
Method	`log_metric`	Log one scalar metric value.
Method	`log_metrics`	Log multiple scalar metrics.
Method	`log_stored_logs_to_mlflow`	Log any data stored in local_logs to MLflow.
Method	`log_to_mlflow`	Log to MLflow with retries.
Method	`save_logs_locally`	Save metrics, parameters, images, artifacts, and matrices locally.
Method	`store_log_locally`	Store log data locally if logging to MLflow fails.
Method	`upload_artifact`	Upload an artifact to the logging service.
Method	`upload_artifact_to_mlflow`	Upload an artifact specifically to MLflow.
Instance Variable	`is_mlflow`	Whether MLflow logging is enabled.
Instance Variable	`is_sweep`	Whether the current run belongs to a sweep.
Instance Variable	`is_tensorboard`	Whether TensorBoard logging is enabled.
Instance Variable	`is_wandb`	Whether WandB logging is enabled.
Instance Variable	`local_logs`	Locally cached MLflow payloads that will be retried or written to disk on close.
Instance Variable	`mlflow_initialized`	Whether MLflow initialization has succeeded.
Instance Variable	`mlflow_tracking_uri`	MLflow tracking URI used when MLflow logging is enabled.
Instance Variable	`project_id`	Project identifier used by WandB and MLflow.
Instance Variable	`project_name`	Project name used by WandB and MLflow.
Instance Variable	`rank`	Process rank. Only rank 0 writes through rank-gated logging methods.
Instance Variable	`run_directory`	Directory for local run artifacts.
Instance Variable	`run_id`	MLflow run identifier, used to resume an existing run.
Instance Variable	`run_name`	Name of the current run.
Instance Variable	`save_directory`	Root directory where local run outputs are stored.
Instance Variable	`wandb_entity`	WandB entity used for logging.
Property	`experiment`	Creates new experiments or returns active ones if already created.
Property	`name`	Run name.
Property	`version`	Tracker version.
Method	`_get_latest_run_name`	Return the most recently created run name.
Method	`_get_next_run_number`	Return the number ID for the next run.
Method	`_get_run_name`	Generate a new run name.
Instance Variable	`_experiment`	Undocumented

@staticmethod

def rank_zero_only(fn: Callable) -> Callable: ¶

Wrap a function so only processes with rank=0 execute it.

def __init__(self, project_name: str | None = None, project_id: str | None = None, run_name: str | None = None, run_id: str | None = None, save_directory: PathType = 'output', is_tensorboard: bool = False, is_wandb: bool = False, is_mlflow: bool = False, is_sweep: bool = False, wandb_entity: str | None = None, mlflow_tracking_uri: str | None = None, rank: int = 0): ¶

Create a tracker for one or more logging integrations.

Parameters
project_name:`str \| None`	Project name used for WandB and MLflow.
project_id:`str \| None`	Project ID used for WandB and MLflow.
run_name:`str \| None`	Run name. If omitted, rank 0 generates a new name and other ranks use the latest run name.
run_id:`str \| None`	MLflow run ID used to continue a previous run.
save_directory:`PathType`	Directory where local outputs are saved.
is_tensorboard:`bool`	Whether to use TensorBoard logging.
is_wandb:`bool`	Whether to use WandB logging.
is_mlflow:`bool`	Whether to use MLflow logging.
is_sweep:`bool`	Whether the current run is part of a sweep.
wandb_entity:`str \| None`	WandB entity to use.
mlflow_tracking_uri:`str \| None`	MLflow tracking URI to use.
rank:`int`	Process rank used in distributed training.

Raises
`ValueError`	If WandB or MLflow is enabled but neither `project_name` nor `project_id` is provided.
`ValueError`	If WandB is enabled without `wandb_entity`.
`ValueError`	If MLflow is enabled without `mlflow_tracking_uri`.
`ValueError`	If no logging integration is enabled.

def close(self): ¶

Finalize logging and save unsent logs locally.

@rank_zero_only

def log_hyperparams(self, params: dict[str, str | bool | int | float | None]): ¶

Log a hyperparameter dictionary.

Parameters
params:`dict[str, str \| bool \| int \| float \| None]`	Hyperparameter key-value pairs.

@rank_zero_only

def log_image(self, name: str, img: np.ndarray, step: int): ¶

Log one image.

Note

step is omitted when logging with WandB to avoid problems with inconsistent incrementation.

Parameters
name:`str`	Image caption. For MLflow, this should include a slash-separated base path and image caption.
img:`np.ndarray`	Image data of shape (H, W, C).
step:`int`	Current step.

@rank_zero_only

def log_images(self, imgs: dict[str, np.ndarray], step: int): ¶

Log multiple images.

Parameters
imgs:`dict[str, np.ndarray]`	Mapping from image captions to image data of shape (H, W, C).
step:`int`	Current step.

@rank_zero_only

def log_matrix(self, matrix: np.ndarray, name: str, step: int, extra_data: dict | None = None): ¶

Log a matrix to the enabled logging services.

Parameters
matrix:`np.ndarray`	Matrix to log, usually of shape (M, N).
name:`str`	Name used for the matrix artifact.
step:`int`	Current step.
extra_data:`dict \| None`	Optional dictionary of additional data to include in the logged matrix artifact.

@rank_zero_only

def log_metric(self, name: str, value: float, step: int): ¶

Log one scalar metric value.

Note

step is omitted when logging with WandB to avoid problems with inconsistent incrementation.

Parameters
name:`str`	Metric name.
value:`float`	Metric value.
step:`int`	Current step.

@rank_zero_only

def log_metrics(self, metrics: dict[str, float], step: int): ¶

Log multiple scalar metrics.

Parameters
metrics:`dict[str, float]`	Metric key-value pairs.
step:`int`	Current step.

def log_stored_logs_to_mlflow(self): ¶

Log any data stored in local_logs to MLflow.

def log_to_mlflow(self, log_fn: Callable, *args, **kwargs): ¶

Log to MLflow with retries.

Logs locally if failures persist.

def save_logs_locally(self): ¶

Save metrics, parameters, images, artifacts, and matrices locally.

def store_log_locally(self, log_fn: Callable, *args, **kwargs): ¶

Store log data locally if logging to MLflow fails.

@rank_zero_only

def upload_artifact(self, path: PathType, name: str | None = None, typ: str = 'artifact'): ¶

Upload an artifact to the logging service.

Parameters
path:`PathType`	Path to the artifact.
name:`str \| None`	Artifact name. If `None`, uses the file stem for WandB and the file name for MLflow.
typ:`str`	The type of the artifact. Only used for WandB.

def upload_artifact_to_mlflow(self, path: PathType, name: str | None = None): ¶

Upload an artifact specifically to MLflow.

Parameters
path:`PathType`	Path to the artifact.
name:`str \| None`	Artifact name. If `None`, uses the file name.

is_mlflow = ¶

Whether MLflow logging is enabled.

is_sweep = ¶

Whether the current run belongs to a sweep.

is_tensorboard = ¶

Whether TensorBoard logging is enabled.

is_wandb = ¶

Whether WandB logging is enabled.

local_logs: dict = ¶

Locally cached MLflow payloads that will be retried or written to disk on close.

mlflow_initialized: bool = ¶

Whether MLflow initialization has succeeded.

mlflow_tracking_uri = ¶

MLflow tracking URI used when MLflow logging is enabled.

project_id = ¶

Project identifier used by WandB and MLflow.

project_name = ¶

Project name used by WandB and MLflow.

rank = ¶

Process rank. Only rank 0 writes through rank-gated logging methods.

run_directory = ¶

Directory for local run artifacts.

run_id = ¶

MLflow run identifier, used to resume an existing run.

run_name = ¶

Name of the current run.

save_directory = ¶

Root directory where local run outputs are stored.

wandb_entity = ¶

WandB entity used for logging.

@property
@rank_zero_only

experiment: dict[Literal['tensorboard', 'wandb', 'mlflow'], Any] = ¶

Creates new experiments or returns active ones if already created.

@property

name: str = ¶

Run name.

Returns
Current run name.

@property

version: int = ¶

Tracker version.

Returns
Version number 1.

def _get_latest_run_name(self) -> str: ¶

Return the most recently created run name.

def _get_next_run_number(self) -> int: ¶

Return the number ID for the next run.

def _get_run_name(self) -> str: ¶

Generate a new run name.

_experiment = ¶

Undocumented