Skip to content

confusion_matrix

Confusion Matrix Metric API

Classes:

  • ConfusionMatrix –

    Accumulates a confusion matrix and returns row-normalized values.

Classes

ConfusionMatrix 

ConfusionMatrix(num_classes: int, name='confusion_matrix', **kwargs)
Source code in helia_edge/metrics/confusion_matrix.py 
16
17
18
19
20
21
22
23
24
25
def __init__(self, num_classes: int, name="confusion_matrix", **kwargs):
    super().__init__(name=name, **kwargs)
    self.num_classes = int(num_classes)
    self._state_dtype = self.dtype or "float32"
    self.conf_matrix = self.add_variable(
        name="conf_matrix",
        shape=(self.num_classes, self.num_classes),
        initializer="zeros",
        dtype=self._state_dtype,
    )

Functions

update_state 
update_state(y_true, y_pred, sample_weight=None)

y_pred: shape (batch, ..., num_classes) or integer class labels y_true: shape (batch, ...) with integer class labels

Source code in helia_edge/metrics/confusion_matrix.py 
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
def update_state(self, y_true, y_pred, sample_weight=None):
    """
    y_pred: shape (batch, ..., num_classes) or integer class labels
    y_true: shape (batch, ...) with integer class labels
    """
    y_true = keras.ops.convert_to_tensor(y_true)
    y_pred = keras.ops.convert_to_tensor(y_pred)

    if y_pred.shape is not None and len(y_pred.shape) > 1 and y_pred.shape[-1] == self.num_classes:
        pred_labels = keras.ops.argmax(y_pred, axis=-1)
    else:
        pred_labels = y_pred

    y_true_flat = keras.ops.cast(keras.ops.reshape(y_true, (-1,)), "int32")
    pred_flat = keras.ops.cast(keras.ops.reshape(pred_labels, (-1,)), "int32")

    if sample_weight is not None:
        sample_weight = keras.ops.convert_to_tensor(sample_weight, dtype=self._state_dtype)
        sample_weight = keras.ops.reshape(sample_weight, (-1,))

    batch_conf_matrix = metrics_utils.confusion_matrix(
        labels=y_true_flat,
        predictions=pred_flat,
        num_classes=self.num_classes,
        weights=sample_weight,
        dtype=self._state_dtype,
    )

    self.conf_matrix.assign_add(batch_conf_matrix)
result 
result()

Returns the row-normalized confusion matrix as float64.

Source code in helia_edge/metrics/confusion_matrix.py 
57
58
59
60
61
62
def result(self):
    """Returns the row-normalized confusion matrix as float64."""
    conf_matrix_f64 = keras.ops.cast(self.conf_matrix, "float64")
    row_sums = keras.ops.sum(conf_matrix_f64, axis=1, keepdims=True)
    normalized = keras.ops.divide(conf_matrix_f64, row_sums)
    return keras.ops.nan_to_num(normalized, posinf=0, neginf=0)