Style updates (#96)

* Bump version: 0.4.2 → 0.4.3

* style

Author: bw4sz

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.4.2
+current_version = 0.4.3
 commit = True
 tag = True
 parse = (?P<major>\d+)\.(?P<minor>\d+)\.(?P<patch>\d+)

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "milliontrees"
-version = "0.4.2"
+version = "0.4.3"
 description = "Benchmark dataset for Airborne Tree Machine Learning"
 readme = "README.md"
 license = { text = "MIT" }

src/milliontrees/common/data_loaders.py

@@ -114,7 +114,8 @@ def get_eval_loader(loader, dataset, batch_size, grouper=None, **loader_kwargs):
 
 
 class GroupSampler:
-    """Constructs batches by first sampling groups, then sampling data from those groups.
+    """Constructs batches by first sampling groups, then sampling data from
+    those groups.
 
     It drops the last batch if it's incomplete.
     """

src/milliontrees/common/grouper.py

@@ -12,8 +12,8 @@
 class Grouper:
     """Groupers group data points together based on their metadata.
 
-    They are used for training and evaluation, e.g., to measure the accuracies of different groups
-    of data.
+    They are used for training and evaluation, e.g., to measure the
+    accuracies of different groups of data.
     """
 
     def __init__(self):

src/milliontrees/common/metrics/all_metrics.py

@@ -43,7 +43,8 @@ def binary_logits_to_pred(logits):
 
 
 def pseudolabel_binary_logits(logits, confidence_threshold):
-    """Applies a confidence threshold to binary logits and generates pseudo- labels.
+    """Applies a confidence threshold to binary logits and generates pseudo-
+    labels.
 
     Args:
         logits (Tensor): A tensor of shape (batch_size, n_tasks) representing binary logits.
@@ -78,7 +79,8 @@ def pseudolabel_binary_logits(logits, confidence_threshold):
 
 
 def pseudolabel_multiclass_logits(logits, confidence_threshold):
-    """Applies a confidence threshold to multi-class logits and generates pseudo-labels.
+    """Applies a confidence threshold to multi-class logits and generates
+    pseudo-labels.
 
     Args:
         logits (Tensor): A tensor of shape (batch_size, ..., n_classes) representing multi-class logits.
@@ -145,7 +147,8 @@ def pseudolabel_detection(preds, confidence_threshold):
 
 
 def pseudolabel_detection_discard_empty(preds, confidence_threshold):
-    """Filters detection predictions based on a confidence threshold and discards empty entries.
+    """Filters detection predictions based on a confidence threshold and
+    discards empty entries.
 
     Args:
         preds (List[dict]): A list of length `batch_size`, where each entry is a dictionary
@@ -355,7 +358,8 @@ def __init__(self, name=None):
 
 
 class PrecisionAtRecall(Metric):
-    """Given a specific model threshold, determine the precision score achieved."""
+    """Given a specific model threshold, determine the precision score
+    achieved."""
 
     def __init__(self, threshold, score_fn=None, name=None):
         self.score_fn = score_fn
@@ -400,8 +404,8 @@ def worst(self, metrics):
 
 
 class DetectionAccuracy(ElementwiseMetric):
-    """Given a specific Intersection over union threshold, determine the accuracy achieved for a
-    one-class detector."""
+    """Given a specific Intersection over union threshold, determine the
+    accuracy achieved for a one-class detector."""
 
     def __init__(self,
                  iou_threshold=0.3,
@@ -488,8 +492,8 @@ def worst(self, metrics):
 
 
 class KeypointAccuracy(ElementwiseMetric):
-    """Given a specific Intersection over union threshold, determine the accuracy achieved for a
-    one-class detector."""
+    """Given a specific Intersection over union threshold, determine the
+    accuracy achieved for a one-class detector."""
 
     def __init__(self,
                  distance_threshold=0.1,
@@ -563,8 +567,8 @@ def worst(self, metrics):
 
 
 class MaskAccuracy(ElementwiseMetric):
-    """Given a specific Intersection over union threshold, determine the accuracy achieved for a
-    Mask R-CNN detector."""
+    """Given a specific Intersection over union threshold, determine the
+    accuracy achieved for a Mask R-CNN detector."""
 
     def __init__(self,
                  iou_threshold=0.5,
@@ -737,8 +741,8 @@ def worst(self, metrics):
 class CountingError(ElementwiseMetric):
     """Mean Absolute Error between ground truth and predicted detection counts.
 
-    Calculates MAE between the number of detections in ground truth vs predictions for each sample
-    in the batch.
+    Calculates MAE between the number of detections in ground truth vs
+    predictions for each sample in the batch.
     """
 
     def __init__(self, score_threshold=0.1, name=None, geometry_name="y"):

src/milliontrees/common/metrics/loss.py

@@ -12,7 +12,8 @@ def __init__(self, loss_fn, name=None):
         super().__init__(name=name)
 
     def _compute(self, y_pred, y_true):
-        """Helper for computing element-wise metric, implemented for each metric.
+        """Helper for computing element-wise metric, implemented for each
+        metric.
 
         Args:
             - y_pred (Tensor): Predicted targets or model output
@@ -23,7 +24,8 @@ def _compute(self, y_pred, y_true):
         return self.loss_fn(y_pred, y_true)
 
     def worst(self, metrics):
-        """Given a list/numpy array/Tensor of metrics, computes the worst-case metric.
+        """Given a list/numpy array/Tensor of metrics, computes the worst-case
+        metric.
 
         Args:
             - metrics (Tensor, numpy array, or list): Metrics
@@ -42,7 +44,8 @@ def __init__(self, loss_fn, name=None):
         super().__init__(name=name)
 
     def _compute_element_wise(self, y_pred, y_true):
-        """Helper for computing element-wise metric, implemented for each metric.
+        """Helper for computing element-wise metric, implemented for each
+        metric.
 
         Args:
             - y_pred (Tensor): Predicted targets or model output
@@ -53,7 +56,8 @@ def _compute_element_wise(self, y_pred, y_true):
         return self.loss_fn(y_pred, y_true)
 
     def worst(self, metrics):
-        """Given a list/numpy array/Tensor of metrics, computes the worst-case metric.
+        """Given a list/numpy array/Tensor of metrics, computes the worst-case
+        metric.
 
         Args:
             - metrics (Tensor, numpy array, or list): Metrics
@@ -81,7 +85,8 @@ def _compute_flattened(self, flattened_y_pred, flattened_y_true):
         return flattened_loss
 
     def worst(self, metrics):
-        """Given a list/numpy array/Tensor of metrics, computes the worst-case metric.
+        """Given a list/numpy array/Tensor of metrics, computes the worst-case
+        metric.
 
         Args:
             - metrics (Tensor, numpy array, or list): Metrics

src/milliontrees/common/metrics/metric.py

@@ -10,7 +10,8 @@ def __init__(self, name):
         self._name = name
 
     def _compute(self, y_pred, y_true):
-        """Helper function for computing the metric. Subclasses should implement this.
+        """Helper function for computing the metric. Subclasses should
+        implement this.
 
         Args:
             - y_pred (Tensor): Predicted targets or model output
@@ -21,7 +22,8 @@ def _compute(self, y_pred, y_true):
         return NotImplementedError
 
     def worst(self, metrics):
-        """Given a list/numpy array/Tensor of metrics, computes the worst-case metric.
+        """Given a list/numpy array/Tensor of metrics, computes the worst-case
+        metric.
 
         Args:
             - metrics (Tensor, numpy array, or list): Metrics
@@ -34,33 +36,35 @@ def worst(self, metrics):
     def name(self):
         """Metric name.
 
-        Used to name the key in the results dictionaries returned by the metric.
+        Used to name the key in the results dictionaries returned by the
+        metric.
         """
         return self._name
 
     @property
     def agg_metric_field(self):
-        """The name of the key in the results dictionary returned by Metric.compute().
+        """The name of the key in the results dictionary returned by
+        Metric.compute().
 
-        This should correspond to the aggregate metric computed on all of y_pred and y_true, in
-        contrast to a group-wise evaluation.
+        This should correspond to the aggregate metric computed on all
+        of y_pred and y_true, in contrast to a group-wise evaluation.
         """
         return f'{self.name}_all'
 
     def group_metric_field(self, group_idx):
-        """The name of the keys corresponding to individual group evaluations in the results
-        dictionary returned by Metric.compute_group_wise()."""
+        """The name of the keys corresponding to individual group evaluations
+        in the results dictionary returned by Metric.compute_group_wise()."""
         return f'{self.name}_group:{group_idx}'
 
     @property
     def worst_group_metric_field(self):
-        """The name of the keys corresponding to the worst-group metric in the results dictionary
-        returned by Metric.compute_group_wise()."""
+        """The name of the keys corresponding to the worst-group metric in the
+        results dictionary returned by Metric.compute_group_wise()."""
         return f'{self.name}_wg'
 
     def group_count_field(self, group_idx):
-        """The name of the keys corresponding to each group's count in the results dictionary
-        returned by Metric.compute_group_wise()."""
+        """The name of the keys corresponding to each group's count in the
+        results dictionary returned by Metric.compute_group_wise()."""
         return f'count_group:{group_idx}'
 
     def compute(self, y_pred, y_true, return_dict=True):
@@ -140,7 +144,8 @@ class ElementwiseMetric(Metric):
     """Averages."""
 
     def _compute_element_wise(self, y_pred, y_true):
-        """Helper for computing element-wise metric, implemented for each metric.
+        """Helper for computing element-wise metric, implemented for each
+        metric.
 
         Args:
             - y_pred (Tensor): Predicted targets or model output
@@ -151,7 +156,8 @@ def _compute_element_wise(self, y_pred, y_true):
         raise NotImplementedError
 
     def worst(self, metrics):
-        """Given a list/numpy array/Tensor of metrics, computes the worst-case metric.
+        """Given a list/numpy array/Tensor of metrics, computes the worst-case
+        metric.
 
         Args:
             - metrics (Tensor, numpy array, or list): Metrics
@@ -182,7 +188,8 @@ def _compute_group_wise(self, y_pred, y_true, g, n_groups):
 
     @property
     def agg_metric_field(self):
-        """The name of the key in the results dictionary returned by Metric.compute()."""
+        """The name of the key in the results dictionary returned by
+        Metric.compute()."""
         return f'{self.name}_avg'
 
     def compute_element_wise(self, y_pred, y_true, return_dict=True):

src/milliontrees/common/utils.py

@@ -40,7 +40,8 @@ def maximum(numbers, empty_val=0.):
 
 
 def split_into_groups(g):
-    """Splits the input tensor into unique groups and their corresponding indices.
+    """Splits the input tensor into unique groups and their corresponding
+    indices.
 
     Args:
         g (Tensor): A vector containing group labels.
@@ -64,9 +65,10 @@ def split_into_groups(g):
 
 
 def get_counts(g, n_groups):
-    """This differs from split_into_groups in how it handles missing groups. get_counts always
-    returns a count array of length n_groups, whereas split_into_groups returns a unique_counts
-    array whose length is the number of unique groups present in g.
+    """This differs from split_into_groups in how it handles missing groups.
+    get_counts always returns a count array of length n_groups, whereas
+    split_into_groups returns a unique_counts array whose length is the number
+    of unique groups present in g.
 
     Args:
         - g (ndarray): Vector of groups
@@ -140,7 +142,8 @@ def shuffle_arr(arr, seed=None):
 
 
 def threshold_at_recall(y_pred, y_true, global_recall=60):
-    """Calculate the model threshold used to achieve a desired global_recall level.
+    """Calculate the model threshold used to achieve a desired global_recall
+    level.
 
     Args:
         y_pred (Description of y_pred, Assumes that y_true is a vector of the true binary labels.)

src/milliontrees/datasets/TreeBoxes.py

@@ -19,7 +19,8 @@
 
 
 class TreeBoxesDataset(MillionTreesDataset):
-    """A dataset of tree annotations with bounding box coordinates from multiple global sources.
+    """A dataset of tree annotations with bounding box coordinates from
+    multiple global sources.
 
     The dataset contains aerial imagery of trees with their corresponding bounding box annotations.
     Each tree is annotated with a 4-point bounding box (x_min, y_min, x_max, y_max).
@@ -225,8 +226,8 @@ def __init__(self,
     def eval(self, y_pred, y_true, metadata):
         """Performs evaluation on the given predictions.
 
-        The main evaluation metric, detection_acc_avg_dom, measures the simple average of the
-        detection accuracies of each domain.
+        The main evaluation metric, detection_acc_avg_dom, measures the
+        simple average of the detection accuracies of each domain.
         """
 
         results = {}
@@ -257,7 +258,8 @@ def eval(self, y_pred, y_true, metadata):
         return results, results_str
 
     def _get_mini_versions_dict(self):
-        """Generate mini versions dict with modified URLs for smaller datasets."""
+        """Generate mini versions dict with modified URLs for smaller
+        datasets."""
         mini_versions = {}
         for version, info in self._versions_dict.items():
             mini_info = info.copy()
@@ -290,7 +292,8 @@ def get_input(self, idx):
 
     @staticmethod
     def _collate_fn(batch):
-        """Collates a batch by stacking `x` (features) and `metadata`, but not `y` (targets).
+        """Collates a batch by stacking `x` (features) and `metadata`, but not
+        `y` (targets).
 
         The batch is initially a tuple of individual data points: (item1, item2, item3, ...).
         After zipping, it transforms into a list of tuples:

src/milliontrees/datasets/TreePoints.py

@@ -17,7 +17,8 @@
 
 
 class TreePointsDataset(MillionTreesDataset):
-    """The TreePoints dataset is a collection of tree annotations annotated as x,y locations.
+    """The TreePoints dataset is a collection of tree annotations annotated as
+    x,y locations.
 
     Dataset Splits:
         - random: For each source, 80% of the data is used for training and 20% for testing.
@@ -186,7 +187,8 @@ def __init__(self,
         super().__init__(root_dir, download, split_scheme)
 
     def _get_mini_versions_dict(self):
-        """Generate mini versions dict with modified URLs for smaller datasets."""
+        """Generate mini versions dict with modified URLs for smaller
+        datasets."""
         mini_versions = {}
         for version, info in self._versions_dict.items():
             mini_info = info.copy()
@@ -204,8 +206,8 @@ def get_annotation_from_filename(self, filename):
         return self._y_array[indices]
 
     def eval(self, y_pred, y_true, metadata):
-        """The main evaluation metric, detection_acc_avg_dom, measures the simple average of the
-        detection accuracies of each domain."""
+        """The main evaluation metric, detection_acc_avg_dom, measures the
+        simple average of the detection accuracies of each domain."""
 
         results = {}
         results_str = ''
@@ -256,8 +258,8 @@ def get_input(self, idx):
     def _collate_fn(batch):
         """Stack x (batch[1]) and metadata (batch[0]), but not y.
 
-        originally, batch = (item1, item2, item3, item4) after zip, batch = [(item1[0], item2[0],
-        ..), ..]
+        originally, batch = (item1, item2, item3, item4) after zip,
+        batch = [(item1[0], item2[0], ..), ..]
         """
         batch = list(zip(*batch))
         batch[1] = torch.stack(batch[1])