Module delta.extensions.metrics

Various helpful loss functions.

Classes

class SparseBinaryAccuracy (label=None, class_id: int = None, name: str = None, binary: int = False, **kwargs)

Accuracy.

Parameters

label

A class identifier accepted by ClassesConfig.class_id(). Compared to valuse in the label image.

class_id : Optional[int]

For multi-class one-hot outputs, used if the output class ID is different than the one in the label image.

label_id : Optional[int]

Internal use only, for reconstructing this class from a .savedmodel format

name : str

Metric name.

binary : bool

Use binary threshold (0.5) (one input, two classes) or argmax on one-hot encoding (one input per class).

Expand source code

class SparseBinaryAccuracy(SparseMetric): # pragma: no cover
    """
    Accuracy.
    """
    def __init__(self, label=None, class_id: int=None, name: str=None, binary: int=False, **kwargs):
        super().__init__(label, class_id, name, binary, **kwargs)
        self._nodata_id = config.dataset.classes.class_id('nodata')
        self._total = self.add_weight('total', initializer='zeros')
        self._correct = self.add_weight('correct', initializer='zeros')

    def update_state(self, y_true, y_pred, sample_weight=None): #pylint: disable=unused-argument, arguments-differ
        y_true = tf.squeeze(y_true)
        right_class = tf.math.equal(y_true, self._label_id)
        if self._binary:
            y_pred = y_pred >= 0.5
            right_class_pred = tf.squeeze(y_pred)
        else:
            y_pred = tf.math.argmax(y_pred, axis=-1)
            right_class_pred = tf.math.equal(y_pred, self._class_id)

        true_positives = tf.math.logical_and(right_class, right_class_pred)
        false_negatives = tf.math.logical_and(tf.math.logical_not(right_class), tf.math.logical_not(right_class_pred))
        if self._nodata_id:
            valid = tf.math.not_equal(y_true, self._nodata_id)
            true_positives = tf.math.logical_and(true_positives, valid)
            false_negatives = tf.math.logical_and(false_negatives, valid)
            total = tf.math.reduce_sum(tf.cast(valid, tf.float32))
        else:
            total = tf.size(y_true)

        true_positives = tf.math.reduce_sum(tf.cast(true_positives, tf.float32))
        false_negatives = tf.math.reduce_sum(tf.cast(false_negatives, tf.float32))
        self._correct.assign_add(true_positives + false_negatives)
        self._total.assign_add(total)

    def result(self):
        return tf.math.divide(self._correct, self._total)

Ancestors

• SparseMetric
• keras.src.metrics.base_metric.Metric
• keras.src.engine.base_layer.Layer
• tensorflow.python.module.module.Module
• tensorflow.python.trackable.autotrackable.AutoTrackable
• tensorflow.python.trackable.base.Trackable
• keras.src.utils.version_utils.LayerVersionSelector

Methods

def result(self)

Computes and returns the scalar metric value tensor or a dict of scalars.

Result computation is an idempotent operation that simply calculates the metric value using the state variables.

Returns

A scalar tensor, or a dictionary of scalar tensors.

def update_state(self, y_true, y_pred, sample_weight=None)

Accumulates statistics for the metric.

Note: This function is executed as a graph function in graph mode. This means: a) Operations on the same resource are executed in textual order. This should make it easier to do things like add the updated value of a variable to another, for example. b) You don't need to worry about collecting the update ops to execute. All update ops added to the graph by this function will be executed. As a result, code should generally work the same way with graph or eager execution.

Args

*args:

**kwargs

A mini-batch of inputs to the Metric.

Inherited members

• SparseMetric:
  • get_config
  • reset_state

class SparseMetric (label=None, class_id: int = None, name: str = None, binary: int = False, label_id: int = None, **kwargs)

An abstract class for metrics applied to integer class labels, with networks that output one-hot encoding.

Parameters

label

A class identifier accepted by ClassesConfig.class_id(). Compared to valuse in the label image.

class_id : Optional[int]

For multi-class one-hot outputs, used if the output class ID is different than the one in the label image.

label_id : Optional[int]

Internal use only, for reconstructing this class from a .savedmodel format

name : str

Metric name.

binary : bool

Use binary threshold (0.5) (one input, two classes) or argmax on one-hot encoding (one input per class).

Expand source code

class SparseMetric(tensorflow.keras.metrics.Metric): # pylint:disable=abstract-method # pragma: no cover
    """
    An abstract class for metrics applied to integer class labels,
    with networks that output one-hot encoding.
    """
    def __init__(self, label=None, class_id: int=None, name: str=None, binary: int=False, label_id: int=None, **kwargs):
        """
        Parameters
        ----------
        label
            A class identifier accepted by `delta.imagery.imagery_config.ClassesConfig.class_id`.
            Compared to valuse in the label image.
        class_id: Optional[int]
            For multi-class one-hot outputs, used if the output class ID is different than the
            one in the label image.
        label_id: Optional[int]
            Internal use only, for reconstructing this class from a .savedmodel format
        name: str
            Metric name.
        binary: bool
            Use binary threshold (0.5) (one input, two classes) or argmax on one-hot encoding (one input per class).
        """
        super().__init__(name=name, **kwargs)
        self._binary = binary
        self._label_id = config.dataset.classes.class_id(label) if label_id is None else label_id
        self._class_id = class_id if class_id is not None else self._label_id

    def reset_state(self):
        for s in self.variables:
            s.assign(tf.zeros(shape=s.shape))

    def get_config(self):
        cfg = super().get_config()
        cfg.update({'binary': self._binary, 'class_id': self._class_id, 'label_id': self._label_id})
        return cfg

Ancestors

• keras.src.metrics.base_metric.Metric
• keras.src.engine.base_layer.Layer
• tensorflow.python.module.module.Module
• tensorflow.python.trackable.autotrackable.AutoTrackable
• tensorflow.python.trackable.base.Trackable
• keras.src.utils.version_utils.LayerVersionSelector

Subclasses

• SparseBinaryAccuracy
• SparsePrecision
• SparseRecall

Methods

def get_config(self)

Returns the serializable config of the metric.

def reset_state(self)

Resets all of the metric state variables.

This function is called between epochs/steps, when a metric is evaluated during training.

class SparsePrecision (label=None, class_id: int = None, name: str = None, binary: int = False, **kwargs)

Precision.

Parameters

label

A class identifier accepted by ClassesConfig.class_id(). Compared to valuse in the label image.

class_id : Optional[int]

For multi-class one-hot outputs, used if the output class ID is different than the one in the label image.

label_id : Optional[int]

Internal use only, for reconstructing this class from a .savedmodel format

name : str

Metric name.

binary : bool

Use binary threshold (0.5) (one input, two classes) or argmax on one-hot encoding (one input per class).

Expand source code

class SparsePrecision(SparseMetric): # pragma: no cover
    """
    Precision.
    """
    def __init__(self, label=None, class_id: int=None, name: str=None, binary: int=False, **kwargs):

        super().__init__(label, class_id, name, binary, **kwargs)
        self._nodata_id = config.dataset.classes.class_id('nodata')
        self._total_class = self.add_weight('total_class', initializer='zeros')
        self._true_positives = self.add_weight('true_positives', initializer='zeros')

    def update_state(self, y_true, y_pred, sample_weight=None): #pylint: disable=unused-argument, arguments-differ
        y_true = tf.squeeze(y_true)
        right_class = tf.math.equal(y_true, self._label_id)
        if self._binary:
            y_pred = y_pred >= 0.5
            right_class_pred = tf.squeeze(y_pred)
        else:
            y_pred = tf.math.argmax(y_pred, axis=-1)
            right_class_pred = tf.math.equal(y_pred, self._class_id)

        if self._nodata_id:
            valid = tf.math.not_equal(y_true, self._nodata_id)
            right_class_pred = tf.math.logical_and(right_class_pred, valid)

        total_class = tf.math.reduce_sum(tf.cast(right_class_pred, tf.float32))
        self._total_class.assign_add(total_class)
        true_positives = tf.math.logical_and(right_class, right_class_pred)
        true_positives = tf.math.reduce_sum(tf.cast(true_positives, tf.float32))
        self._true_positives.assign_add(true_positives)

    def result(self):
        return tf.math.divide_no_nan(self._true_positives, self._total_class)

Ancestors

• SparseMetric
• keras.src.metrics.base_metric.Metric
• keras.src.engine.base_layer.Layer
• tensorflow.python.module.module.Module
• tensorflow.python.trackable.autotrackable.AutoTrackable
• tensorflow.python.trackable.base.Trackable
• keras.src.utils.version_utils.LayerVersionSelector

Methods

def result(self)

Computes and returns the scalar metric value tensor or a dict of scalars.

Result computation is an idempotent operation that simply calculates the metric value using the state variables.

Returns

A scalar tensor, or a dictionary of scalar tensors.

def update_state(self, y_true, y_pred, sample_weight=None)

Accumulates statistics for the metric.

Note: This function is executed as a graph function in graph mode. This means: a) Operations on the same resource are executed in textual order. This should make it easier to do things like add the updated value of a variable to another, for example. b) You don't need to worry about collecting the update ops to execute. All update ops added to the graph by this function will be executed. As a result, code should generally work the same way with graph or eager execution.

Args

*args:

**kwargs

A mini-batch of inputs to the Metric.

Inherited members

• SparseMetric:
  • get_config
  • reset_state

class SparseRecall (label=None, class_id: int = None, name: str = None, binary: int = False, **kwargs)

Recall.

Parameters

label

A class identifier accepted by ClassesConfig.class_id(). Compared to valuse in the label image.

class_id : Optional[int]

For multi-class one-hot outputs, used if the output class ID is different than the one in the label image.

label_id : Optional[int]

Internal use only, for reconstructing this class from a .savedmodel format

name : str

Metric name.

binary : bool

Use binary threshold (0.5) (one input, two classes) or argmax on one-hot encoding (one input per class).

Expand source code

class SparseRecall(SparseMetric): # pragma: no cover
    """
    Recall.
    """
    def __init__(self, label=None, class_id: int=None, name: str=None, binary: int=False, **kwargs):

        super().__init__(label, class_id, name, binary, **kwargs)
        self._total_class = self.add_weight('total_class', initializer='zeros')
        self._true_positives = self.add_weight('true_positives', initializer='zeros')

    # sample_weight is unused but required by tensorflow
    def update_state(self, y_true, y_pred, sample_weight=None): #pylint: disable=unused-argument, arguments-differ
        y_true = tf.squeeze(y_true)
        right_class = tf.math.equal(y_true, self._label_id)
        if self._binary:
            y_pred = y_pred >= 0.5
            right_class_pred = tf.squeeze(y_pred)
        else:
            y_pred = tf.math.argmax(y_pred, axis=-1)
            right_class_pred = tf.math.equal(y_pred, self._class_id)
        total_class = tf.math.reduce_sum(tf.cast(right_class, tf.float32))
        self._total_class.assign_add(total_class)
        true_positives = tf.math.logical_and(right_class, right_class_pred)
        true_positives = tf.math.reduce_sum(tf.cast(true_positives, tf.float32))
        self._true_positives.assign_add(true_positives)

    def result(self):
        return tf.math.divide_no_nan(self._true_positives, self._total_class)

Ancestors

• SparseMetric
• keras.src.metrics.base_metric.Metric
• keras.src.engine.base_layer.Layer
• tensorflow.python.module.module.Module
• tensorflow.python.trackable.autotrackable.AutoTrackable
• tensorflow.python.trackable.base.Trackable
• keras.src.utils.version_utils.LayerVersionSelector

Methods

def result(self)

Computes and returns the scalar metric value tensor or a dict of scalars.

Result computation is an idempotent operation that simply calculates the metric value using the state variables.

Returns

A scalar tensor, or a dictionary of scalar tensors.

def update_state(self, y_true, y_pred, sample_weight=None)

Accumulates statistics for the metric.

Note: This function is executed as a graph function in graph mode. This means: a) Operations on the same resource are executed in textual order. This should make it easier to do things like add the updated value of a variable to another, for example. b) You don't need to worry about collecting the update ops to execute. All update ops added to the graph by this function will be executed. As a result, code should generally work the same way with graph or eager execution.

Args

*args:

**kwargs

A mini-batch of inputs to the Metric.

Inherited members

• SparseMetric:
  • get_config
  • reset_state