Skip to content

deeplabcut.pose_estimation_pytorch.runners.logger

Classes:

Name Description
BaseLogger

Base class for logging training runs.
CSVLogger

Logger saving stats and metrics to a CSV file.
ImageLoggerMixin

Mixin for loggers that can log images.
WandbLogger

Wandb logger to track experiments and log data.

Functions:

Name Description
destroy_file_logging

Resets the logging module to log everything to the console.
setup_file_logging

Sets up logging to a file.

BaseLogger

Bases: ABC

Base class for logging training runs.

Methods:

Name Description
log

Logs data from a training run.
log_config

Logs the configuration data for a training run.
save

Saves the current training logs.
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
class BaseLogger(ABC):
    """Base class for logging training runs."""

    @abstractmethod
    def log_config(self, config: dict = None) -> None:
        """Logs the configuration data for a training run.

        Args:
            config: the training configuration used for the run
        """

    @abstractmethod
    def log(self, metrics: dict[str, Any], step: int | None = None) -> None:
        """Logs data from a training run.

        Args:
            metrics: the metrics to log
            step: The global step in processing. Defaults to None.
        """

    @abstractmethod
    def save(self) -> None:
        """Saves the current training logs."""

log abstractmethod 

log(metrics: dict[str, Any], step: int | None = None) -> None

Logs data from a training run.

Parameters:

Name Type Description Default

metrics

dict[str, Any]

the metrics to log

 required

step

int | None

The global step in processing. Defaults to None.

 None
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
81
82
83
84
85
86
87
88
@abstractmethod
def log(self, metrics: dict[str, Any], step: int | None = None) -> None:
    """Logs data from a training run.

    Args:
        metrics: the metrics to log
        step: The global step in processing. Defaults to None.
    """

log_config abstractmethod 

log_config(config: dict = None) -> None

Logs the configuration data for a training run.

Parameters:

Name Type Description Default

config

dict

the training configuration used for the run

 None
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
73
74
75
76
77
78
79
@abstractmethod
def log_config(self, config: dict = None) -> None:
    """Logs the configuration data for a training run.

    Args:
        config: the training configuration used for the run
    """

save abstractmethod 

save() -> None

Saves the current training logs.

Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
90
91
92
@abstractmethod
def save(self) -> None:
    """Saves the current training logs."""

CSVLogger

Bases: BaseLogger

Logger saving stats and metrics to a CSV file.

Methods:

Name Description
__init__

Initialize the CSVLogger class.
log

Logs metrics from runs.
log_config

Does not do anything as the config should already be saved.
save

Saves the metrics to the file system.
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
@LOGGER.register_module
class CSVLogger(BaseLogger):
    """Logger saving stats and metrics to a CSV file."""

    def __init__(self, train_folder: str, log_filename: str) -> None:
        """Initialize the CSVLogger class.

        Args:
            train_folder: The path of the folder containing training files.
            log_filename: The name of the file in which to store training stats
        """
        super().__init__()
        train_folder = Path(train_folder)
        self.train_folder = train_folder
        self.log_filename = log_filename
        self.log_file = train_folder / log_filename

        self._steps: list[int] = []
        self._metric_store: list[dict] = []
        self._logged_metrics: set[str] = set()

        # Load existing data if the file exists (e.g., when resuming from snapshot)
        if self.log_file.exists():
            self._load_existing_data()

    def log(self, metrics: dict[str, Any], step: int | None = None) -> None:
        """Logs metrics from runs.

        Args:
            metrics: the metrics to log
            step: The global step in processing. Defaults to None.
        """
        if step is None:
            if len(self._steps) == 0:
                step = 0
            else:
                step = self._steps[-1] + 1

        self._logged_metrics = self._logged_metrics.union(metrics.keys())
        if len(self._steps) > 0 and step == self._steps[-1]:
            self._metric_store[-1].update(metrics)
        else:
            self._steps.append(step)
            self._metric_store.append(metrics)

        self.save()

    def save(self):
        """Saves the metrics to the file system."""
        logs = self._prepare_logs()
        with open(self.log_file, "w", newline="") as f:
            writer = csv.writer(f)
            writer.writerows(logs)

    def log_config(self, config: dict = None) -> None:
        """Does not do anything as the config should already be saved.

        Args:
            config: Experiment config file.
        """
        pass

    def _load_existing_data(self) -> None:
        """Loads existing CSV data if the log file exists."""
        logging.info(f"Loading existing CSV data from {self.log_file}")
        try:
            with open(self.log_file, newline="") as f:
                reader = csv.DictReader(f)

                # Update logged metrics from header
                if "step" not in reader.fieldnames:
                    raise ValueError("Invalid CSV format: missing 'step' column")

                metric_names = [m for m in reader.fieldnames if m != "step"]
                self._logged_metrics.update(metric_names)

                # Load data rows
                steps = []
                metric_store = []
                for row in reader:
                    try:
                        step = int(row["step"])
                    except (ValueError, KeyError):
                        logging.warning(f"Invalid step value in row: {row}")
                        continue

                    # Convert metric values: empty strings -> None, numeric strings -> float
                    step_metrics = {}
                    for metric in metric_names:
                        value = row.get(metric, "").strip()
                        if not value:
                            step_metrics[metric] = None
                        else:
                            try:
                                step_metrics[metric] = float(value)
                            except ValueError:
                                step_metrics[metric] = value

                    steps.append(step)
                    metric_store.append(step_metrics)

        except Exception as e:
            logging.warning(f"Failed to load existing CSV data from {self.log_file}: {e}. Starting with empty log.")
            return
        self._steps.extend(steps)
        self._metric_store.extend(metric_store)

    def _prepare_logs(self) -> list[list]:
        """Prepares the data to log as a list of strings."""
        if len(self._metric_store) == 0:
            return []

        metrics = list(sorted(self._logged_metrics))
        logs = [["step"] + metrics]
        for step, step_metrics in zip(self._steps, self._metric_store, strict=False):
            # Convert None values to empty strings for proper CSV formatting
            row = [step] + ["" if step_metrics.get(m) is None else step_metrics.get(m) for m in metrics]
            logs.append(row)

        return logs

__init__ 

__init__(train_folder: str, log_filename: str) -> None

Initialize the CSVLogger class.

Parameters:

Name Type Description Default

train_folder

str

The path of the folder containing training files.

 required

log_filename

str

The name of the file in which to store training stats

 required
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
def __init__(self, train_folder: str, log_filename: str) -> None:
    """Initialize the CSVLogger class.

    Args:
        train_folder: The path of the folder containing training files.
        log_filename: The name of the file in which to store training stats
    """
    super().__init__()
    train_folder = Path(train_folder)
    self.train_folder = train_folder
    self.log_filename = log_filename
    self.log_file = train_folder / log_filename

    self._steps: list[int] = []
    self._metric_store: list[dict] = []
    self._logged_metrics: set[str] = set()

    # Load existing data if the file exists (e.g., when resuming from snapshot)
    if self.log_file.exists():
        self._load_existing_data()

log 

log(metrics: dict[str, Any], step: int | None = None) -> None

Logs metrics from runs.

Parameters:

Name Type Description Default

metrics

dict[str, Any]

the metrics to log

 required

step

int | None

The global step in processing. Defaults to None.

 None
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
def log(self, metrics: dict[str, Any], step: int | None = None) -> None:
    """Logs metrics from runs.

    Args:
        metrics: the metrics to log
        step: The global step in processing. Defaults to None.
    """
    if step is None:
        if len(self._steps) == 0:
            step = 0
        else:
            step = self._steps[-1] + 1

    self._logged_metrics = self._logged_metrics.union(metrics.keys())
    if len(self._steps) > 0 and step == self._steps[-1]:
        self._metric_store[-1].update(metrics)
    else:
        self._steps.append(step)
        self._metric_store.append(metrics)

    self.save()

log_config 

log_config(config: dict = None) -> None

Does not do anything as the config should already be saved.

Parameters:

Name Type Description Default

config

dict

Experiment config file.

 None
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
442
443
444
445
446
447
448
def log_config(self, config: dict = None) -> None:
    """Does not do anything as the config should already be saved.

    Args:
        config: Experiment config file.
    """
    pass

save 

save()

Saves the metrics to the file system.

Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
435
436
437
438
439
440
def save(self):
    """Saves the metrics to the file system."""
    logs = self._prepare_logs()
    with open(self.log_file, "w", newline="") as f:
        writer = csv.writer(f)
        writer.writerows(logs)

ImageLoggerMixin

Bases: ABC

Mixin for loggers that can log images.

Before starting training, you should call select_images_to_log, which will select a train and a test image for which inputs/outputs will always be logged. Then logger.log_images should be called at every step - the logger will check if anything needs to be uploaded, and take care of it.

Example

project_name = "example" run_name = "run-1" logger = WandbLogger(project_name, run_name) logger.select_images_to_log(train_loader, test_loader)

for i in range(epochs): for batch_inputs in train_loader: batch_labels = batch_data["annotations"] batch_inputs = batch_data["image"] batch_outputs = model(batch_inputs) batch_targets = model.get_target(batch_outputs, batch_labels) loss = criterion(batch_targets, batch_outputs) loss.backwards() optim.step()

    logger.log_images(batch_inputs, batch_outputs, batch_targets)

for batch_inputs in train_loader:
    ...
    logger.log_images(batch_inputs, batch_outputs, batch_targets)

Methods:

Name Description
__init__
log_images

Log images for a batch.
select_images_to_log

Selects the train and test images to log.
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
class ImageLoggerMixin(ABC):
    """Mixin for loggers that can log images.

    Before starting training, you should call `select_images_to_log`, which will
    select a train and a test image for which inputs/outputs will always be logged.
    Then logger.log_images should be called at every step - the logger will check if
    anything needs to be uploaded, and take care of it.

    Example:
        project_name = "example"
        run_name = "run-1"
        logger = WandbLogger(project_name, run_name)
        logger.select_images_to_log(train_loader, test_loader)

        for i in range(epochs):
            for batch_inputs in train_loader:
                batch_labels = batch_data["annotations"]
                batch_inputs = batch_data["image"]
                batch_outputs = model(batch_inputs)
                batch_targets = model.get_target(batch_outputs, batch_labels)
                loss = criterion(batch_targets, batch_outputs)
                loss.backwards()
                optim.step()

                logger.log_images(batch_inputs, batch_outputs, batch_targets)

            for batch_inputs in train_loader:
                ...
                logger.log_images(batch_inputs, batch_outputs, batch_targets)
    """

    def __init__(self, image_log_interval: int | None = None, *args, **kwargs):
        """"""
        super().__init__(*args, **kwargs)
        self.image_log_interval = image_log_interval
        self._logged = {}
        self._denormalize = transforms.Compose(
            [
                transforms.Normalize(mean=[0, 0, 0], std=[1 / 0.229, 1 / 0.224, 1 / 0.225]),
                transforms.Normalize(mean=[-0.485, -0.456, -0.406], std=[1, 1, 1]),
            ]
        )
        self._softmax = torch.nn.Softmax2d()

    @abstractmethod
    def log_images(
        self,
        inputs: dict[str, Any],
        outputs: dict[str, torch.Tensor],
        targets: dict[str, dict[str, torch.Tensor]],
        step: int,
    ) -> None:
        """Log images for a batch.

        Args:
            inputs: the inputs for the model, containing at least an "image" key
            outputs: the outputs of each model head
            targets: the targets for each model head
            step: the current step
        """
        pass

    def select_images_to_log(self, train: DataLoader, valid: DataLoader) -> None:
        """Selects the train and test images to log.

        Args:
            train: the training dataloader
            valid: the inference dataloader
        """

        def _caption(image_path: str) -> str:
            p = Path(image_path)
            return f"{p.parent.name}.{p.stem}"

        train_image = train.dataset[0]["path"]
        test_image = valid.dataset[0]["path"]
        self._logged = {
            train_image: {"name": "train-0", "caption": _caption(train_image)},
            test_image: {"name": "test-0", "caption": _caption(test_image)},
        }

    def _prepare_image(
        self,
        image: torch.Tensor,
        denormalize: bool = False,
        keypoints: torch.Tensor | None = None,
        bboxes: torch.Tensor | None = None,
    ) -> np.ndarray:
        """
        Args:
            image: the image to log, of shape (C, H, W), of any data type
            denormalize: whether to remove ImageNet channel normalization
            keypoints: size (num_instances, K, 2) the K keypoints location
            bboxes: size (N, 4) containing bboxes in (xmin, ymin, xmax, ymax)

        Returns:
            an uint8 array with keypoints and bounding boxes drawn
        """
        if denormalize:
            image = self._denormalize(image.unsqueeze(0)).squeeze()

        image = F.convert_image_dtype(image.detach().cpu(), dtype=torch.uint8)
        if keypoints is not None and len(keypoints) > 0:
            assert len(keypoints.shape) == 3
            # Use visibility and force torchvision >= 0.18
            # pytorch.org/vision/0.18/generated/torchvision.utils.draw_keypoints.html
            # pytorch.org/vision/0.17/generated/torchvision.utils.draw_keypoints.html
            keypoints[torch.any(torch.isnan(keypoints), dim=-1)] = -1
            image = draw_keypoints(image, keypoints=keypoints[..., :2], colors="red", radius=5)

        if bboxes is not None and len(bboxes) > 0:
            assert len(bboxes.shape) == 2
            image = draw_bounding_boxes(image, boxes=bboxes[:, :4], width=1)

        return image.permute(1, 2, 0).numpy()

    def _heatmap_softmax(self, heatmaps: torch.Tensor) -> torch.Tensor:
        """Applies a softmax to the heatmap channels."""
        return self._softmax(heatmaps.detach().cpu())

    def _prepare_images(
        self,
        inputs: dict[str, Any],
        outputs: dict[str, dict[str, torch.Tensor]],
        targets: dict[str, dict[str, dict[str, torch.Tensor]]],
    ) -> dict[str, np.ndarray]:
        """Prepares images for logging."""
        image_logs = {}
        paths = inputs["path"]
        images_to_log = [(i, p) for i, p in enumerate(paths) if p in self._logged]
        for idx, path in images_to_log:
            base = self._logged[path]["name"]
            keypoints = inputs.get("annotations", {}).get("keypoints")
            if keypoints is not None:
                keypoints = keypoints[idx]
            image_logs[f"{base}.input"] = self._prepare_image(
                inputs["image"][idx],
                keypoints=keypoints,
                denormalize=True,
            )

            for head, head_outputs in outputs.items():
                if "heatmap" in head_outputs:
                    head_heatmaps = self._heatmap_softmax(head_outputs["heatmap"][idx])
                    head_targets = targets[head]["heatmap"]["target"][idx]
                    for j, (h, t) in enumerate(zip(head_heatmaps, head_targets, strict=False)):
                        h = self._prepare_image(h.unsqueeze(0))
                        t = self._prepare_image(t.unsqueeze(0))
                        image_logs[f"{base}.heatmap.{j}"] = np.concatenate([h, t])

        return image_logs

__init__ 

__init__(image_log_interval: int | None = None, *args, **kwargs)
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
126
127
128
129
130
131
132
133
134
135
136
137
def __init__(self, image_log_interval: int | None = None, *args, **kwargs):
    """"""
    super().__init__(*args, **kwargs)
    self.image_log_interval = image_log_interval
    self._logged = {}
    self._denormalize = transforms.Compose(
        [
            transforms.Normalize(mean=[0, 0, 0], std=[1 / 0.229, 1 / 0.224, 1 / 0.225]),
            transforms.Normalize(mean=[-0.485, -0.456, -0.406], std=[1, 1, 1]),
        ]
    )
    self._softmax = torch.nn.Softmax2d()

log_images abstractmethod 

log_images(
    inputs: dict[str, Any], outputs: dict[str, Tensor], targets: dict[str, dict[str, Tensor]], step: int
) -> None

Log images for a batch.

Parameters:

Name Type Description Default

inputs

dict[str, Any]

the inputs for the model, containing at least an "image" key

 required

outputs

dict[str, Tensor]

the outputs of each model head

 required

targets

dict[str, dict[str, Tensor]]

the targets for each model head

 required

step

int

the current step

 required
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
@abstractmethod
def log_images(
    self,
    inputs: dict[str, Any],
    outputs: dict[str, torch.Tensor],
    targets: dict[str, dict[str, torch.Tensor]],
    step: int,
) -> None:
    """Log images for a batch.

    Args:
        inputs: the inputs for the model, containing at least an "image" key
        outputs: the outputs of each model head
        targets: the targets for each model head
        step: the current step
    """
    pass

select_images_to_log 

select_images_to_log(train: DataLoader, valid: DataLoader) -> None

Selects the train and test images to log.

Parameters:

Name Type Description Default

train

DataLoader

the training dataloader

 required

valid

DataLoader

the inference dataloader

 required
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
def select_images_to_log(self, train: DataLoader, valid: DataLoader) -> None:
    """Selects the train and test images to log.

    Args:
        train: the training dataloader
        valid: the inference dataloader
    """

    def _caption(image_path: str) -> str:
        p = Path(image_path)
        return f"{p.parent.name}.{p.stem}"

    train_image = train.dataset[0]["path"]
    test_image = valid.dataset[0]["path"]
    self._logged = {
        train_image: {"name": "train-0", "caption": _caption(train_image)},
        test_image: {"name": "test-0", "caption": _caption(test_image)},
    }

WandbLogger

Bases: ImageLoggerMixin, BaseLogger

Wandb logger to track experiments and log data.

Refer to: https://docs.wandb.ai/guides for more information on wandb.

Attributes:

Name Type Description
run Run

The wandb run object associated with the current experiment.

Methods:

Name Description
__init__

Initialize the WandbLogger class.
log

Logs metrics from runs.
log_config

Updates the current run with the given config dict.
log_images

Log images for a batch.
save

Syncs all files to wandb with the policy specified.
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
@LOGGER.register_module
class WandbLogger(ImageLoggerMixin, BaseLogger):
    """Wandb logger to track experiments and log data.

    Refer to: https://docs.wandb.ai/guides for more information on wandb.

    Attributes:
        run (wandb.Run): The wandb run object associated with the current experiment.
    """

    def __init__(
        self,
        project_name: str = "deeplabcut",
        run_name: str = "tmp",
        image_log_interval: int | None = None,
        model: PoseModel = None,
        train_folder: str = None,
        **wandb_kwargs,
    ) -> None:
        """Initialize the WandbLogger class.

        Args:
            project_name: The name of the wandb project. Defaults to "deeplabcut".
            run_name: The name of the wandb run. Defaults to "tmp".
            image_log_interval: How often train/test images are logged in epochs (if
                None, train/test inputs are never logged).
            model: The model to log. Defaults to None.
            train_folder: path to the train folder (used to store the W&B run identifiers)
            wandb_kwargs: extra arguments to pass to ``wb.init``

        Example:
            logger = WandbLogger(project_name="mice", run_name="exp1", model=my_model)
        """
        super().__init__(image_log_interval=image_log_interval)

        if not has_wandb:
            raise ValueError(
                "Cannot use ``WandbLogger`` as wandb is not installed. Please run"
                "``pip install wandb`` if you want to log to wandb"
            )

        if wandb.run is not None:
            wandb.finish()

        self.run = wandb.init(
            project=project_name,
            name=run_name,
            **wandb_kwargs,
        )
        if model is None:
            raise ValueError("Specify the model to track!")
        self.run.watch(model)
        if train_folder is None:
            raise ValueError("Specify the train folder!")
        self.train_folder = Path(train_folder)
        self._save_wandb_info()

    def _save_wandb_info(self):
        wandb_info = {
            "entity": self.run.entity,
            "project": self.run.project,
            "run_id": self.run.id,
        }

        output_path = self.train_folder / "wandb_info.yaml"
        with open(output_path, "w") as f:
            yaml.safe_dump(wandb_info, f)

        logging.info(f"WandB run info saved to {output_path}")

    def log(self, metrics: dict[str, Any], step: int | None = None) -> None:
        """Logs metrics from runs.

        Args:
            metrics: the metrics to log
            step: The global step in processing. Defaults to None.

        Example:
            logger = WandbLogger()
            logger.log({"loss": 0.123}, step=100)
        """
        self.run.log(metrics, step=step)

    def log_images(
        self,
        inputs: dict[str, Any],
        outputs: dict[str, dict[str, torch.Tensor]],
        targets: dict[str, dict[str, dict[str, torch.Tensor]]],
        step: int,
    ) -> None:
        """Log images for a batch.

        Args:
            inputs: the inputs for the model, containing at least an "image" key
            outputs: the outputs of each model head
            targets: the targets for each model head
            step: the current step
        """
        if self.image_log_interval is None or step % self.image_log_interval != 0:
            return

        images = self._prepare_images(inputs, outputs, targets)
        if len(images) > 0:
            self.run.log(
                {name: wandb.Image(image) for name, image in images.items()},
                step=step,
            )

    def save(self):
        """Syncs all files to wandb with the policy specified.

        Notes:
            self.run: A run is a unit of computation logged by wandb.
            self.run.run.dir: The directory where files associated with the run are saved.

        Example:
            logger = WandbLogger()
            # Training and logging
            logger.save()
        """
        self.run.save(self.run.dir)

    def log_config(self, config: dict = None) -> None:
        """Updates the current run with the given config dict.

        Notes:
            self.run: A run is a unit of computation logged by wandb.
            self.run.config: Config object associated with this run.

        Args:
            config: Experiment config file.

        Example:
            logger = WandbLogger()
            config = {"learning_rate": 0.001, "batch_size": 32}
            logger.log_config(config)
        """
        self.run.config.update(config)

__init__ 

__init__(
    project_name: str = "deeplabcut",
    run_name: str = "tmp",
    image_log_interval: int | None = None,
    model: PoseModel = None,
    train_folder: str = None,
    **wandb_kwargs
) -> None

Initialize the WandbLogger class.

Parameters:

Name Type Description Default

project_name

str

The name of the wandb project. Defaults to "deeplabcut".

 'deeplabcut'

run_name

str

The name of the wandb run. Defaults to "tmp".

 'tmp'

image_log_interval

int | None

How often train/test images are logged in epochs (if None, train/test inputs are never logged).

 None

model

PoseModel

The model to log. Defaults to None.

 None

train_folder

str

path to the train folder (used to store the W&B run identifiers)

 None

wandb_kwargs

extra arguments to pass to wb.init

 {}
Example

logger = WandbLogger(project_name="mice", run_name="exp1", model=my_model)

Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
def __init__(
    self,
    project_name: str = "deeplabcut",
    run_name: str = "tmp",
    image_log_interval: int | None = None,
    model: PoseModel = None,
    train_folder: str = None,
    **wandb_kwargs,
) -> None:
    """Initialize the WandbLogger class.

    Args:
        project_name: The name of the wandb project. Defaults to "deeplabcut".
        run_name: The name of the wandb run. Defaults to "tmp".
        image_log_interval: How often train/test images are logged in epochs (if
            None, train/test inputs are never logged).
        model: The model to log. Defaults to None.
        train_folder: path to the train folder (used to store the W&B run identifiers)
        wandb_kwargs: extra arguments to pass to ``wb.init``

    Example:
        logger = WandbLogger(project_name="mice", run_name="exp1", model=my_model)
    """
    super().__init__(image_log_interval=image_log_interval)

    if not has_wandb:
        raise ValueError(
            "Cannot use ``WandbLogger`` as wandb is not installed. Please run"
            "``pip install wandb`` if you want to log to wandb"
        )

    if wandb.run is not None:
        wandb.finish()

    self.run = wandb.init(
        project=project_name,
        name=run_name,
        **wandb_kwargs,
    )
    if model is None:
        raise ValueError("Specify the model to track!")
    self.run.watch(model)
    if train_folder is None:
        raise ValueError("Specify the train folder!")
    self.train_folder = Path(train_folder)
    self._save_wandb_info()

log 

log(metrics: dict[str, Any], step: int | None = None) -> None

Logs metrics from runs.

Parameters:

Name Type Description Default

metrics

dict[str, Any]

the metrics to log

 required

step

int | None

The global step in processing. Defaults to None.

 None
Example

logger = WandbLogger() logger.log({"loss": 0.123}, step=100)

Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
318
319
320
321
322
323
324
325
326
327
328
329
def log(self, metrics: dict[str, Any], step: int | None = None) -> None:
    """Logs metrics from runs.

    Args:
        metrics: the metrics to log
        step: The global step in processing. Defaults to None.

    Example:
        logger = WandbLogger()
        logger.log({"loss": 0.123}, step=100)
    """
    self.run.log(metrics, step=step)

log_config 

log_config(config: dict = None) -> None

Updates the current run with the given config dict.

Notes

self.run: A run is a unit of computation logged by wandb. self.run.config: Config object associated with this run.

Parameters:

Name Type Description Default

config

dict

Experiment config file.

 None
Example

logger = WandbLogger() config = {"learning_rate": 0.001, "batch_size": 32} logger.log_config(config)

Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
def log_config(self, config: dict = None) -> None:
    """Updates the current run with the given config dict.

    Notes:
        self.run: A run is a unit of computation logged by wandb.
        self.run.config: Config object associated with this run.

    Args:
        config: Experiment config file.

    Example:
        logger = WandbLogger()
        config = {"learning_rate": 0.001, "batch_size": 32}
        logger.log_config(config)
    """
    self.run.config.update(config)

log_images 

log_images(
    inputs: dict[str, Any],
    outputs: dict[str, dict[str, Tensor]],
    targets: dict[str, dict[str, dict[str, Tensor]]],
    step: int,
) -> None

Log images for a batch.

Parameters:

Name Type Description Default

inputs

dict[str, Any]

the inputs for the model, containing at least an "image" key

 required

outputs

dict[str, dict[str, Tensor]]

the outputs of each model head

 required

targets

dict[str, dict[str, dict[str, Tensor]]]

the targets for each model head

 required

step

int

the current step

 required
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
def log_images(
    self,
    inputs: dict[str, Any],
    outputs: dict[str, dict[str, torch.Tensor]],
    targets: dict[str, dict[str, dict[str, torch.Tensor]]],
    step: int,
) -> None:
    """Log images for a batch.

    Args:
        inputs: the inputs for the model, containing at least an "image" key
        outputs: the outputs of each model head
        targets: the targets for each model head
        step: the current step
    """
    if self.image_log_interval is None or step % self.image_log_interval != 0:
        return

    images = self._prepare_images(inputs, outputs, targets)
    if len(images) > 0:
        self.run.log(
            {name: wandb.Image(image) for name, image in images.items()},
            step=step,
        )

save 

save()

Syncs all files to wandb with the policy specified.

Notes

self.run: A run is a unit of computation logged by wandb. self.run.run.dir: The directory where files associated with the run are saved.

Example

logger = WandbLogger()

Training and logging

logger.save()

Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
356
357
358
359
360
361
362
363
364
365
366
367
368
def save(self):
    """Syncs all files to wandb with the policy specified.

    Notes:
        self.run: A run is a unit of computation logged by wandb.
        self.run.run.dir: The directory where files associated with the run are saved.

    Example:
        logger = WandbLogger()
        # Training and logging
        logger.save()
    """
    self.run.save(self.run.dir)

destroy_file_logging 

destroy_file_logging() -> None

Resets the logging module to log everything to the console.

Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
62
63
64
65
66
67
def destroy_file_logging() -> None:
    """Resets the logging module to log everything to the console."""
    root = logging.getLogger()
    handlers = [h for h in root.handlers]
    for handler in handlers:
        root.removeHandler(handler)

setup_file_logging 

setup_file_logging(filepath: Path) -> None

Sets up logging to a file.

Parameters:

Name Type Description Default

filepath

Path

the path where logs should be saved

 required
Source code in deeplabcut/pose_estimation_pytorch/runners/logger.py 
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
def setup_file_logging(filepath: Path) -> None:
    """Sets up logging to a file.

    Args:
        filepath: the path where logs should be saved
    """
    logging.basicConfig(
        filename=filepath,
        filemode="a",
        datefmt="%Y-%m-%d %H:%M:%S",
        level=logging.INFO,
        format="%(asctime)-15s %(message)s",
        force=True,
    )
    console_logger = logging.StreamHandler()
    console_logger.setLevel(logging.INFO)
    root = logging.getLogger()
    root.addHandler(console_logger)