# logging

Python API: `luxonis_train.utils.logging`

Configure the global logger through `luxonis_ml`.

## Functions

### setup_logging

```python
def setup_logging(*, file: PathType | None = None, use_rich: bool = True):
```

Set up the global logger for the package.

The function calls `luxonis_ml.utils.setup_logging` with `file` and `use_rich`. That function does these steps:

 * It reads the log level from the `LOG_LEVEL` environment variable. It raises a `ValueError` for an invalid level.
 * It removes the existing `loguru` handlers.
 * It adds a console handler. This handler skips a log record with a `file_only` extra, for example a record from
   `logger.bind(file_only=True)`.
 * It adds a file handler when `file` is not `None`. This handler also writes the records with a `file_only` extra.
 * It sends Python warnings to the logger.
 * It installs an exception hook that prints a summary of an uncaught `pydantic` validation error. The environment variable
   `LUXONISML_DISABLE_PRETTY_VALIDATION_ERRORS` turns the hook off.

When `use_rich` is `True`, the rich tracebacks of logged exceptions show no source code for the frames of `lightning.pytorch`,
`torch`, `pydantic`, and `numpy`. They still show the file and the line of these frames.

The package calls this function when it loads its modules. The config calls it again when it validates `rich_logging`.
[LuxonisModel](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/core/core.md)
calls it with the log file of the run.

Parameters

 * `file` (`PathType | None`): The path of the log file. The logger appends to the file. `None` writes no log file.
 * `use_rich` (`bool`): When `True`, write the console output with a rich handler to `stdout`. When `False`, write plain log lines
   to `stderr`.