ADLStream.evaluation.metrics

Metrics factory.

accuracy(y, o)

Accuracy score.

Parameters:

Name	Type	Description	Default
y	np.array	Real values.	required
o	np.array	Predictions.	required

Returns:

Type	Description
float	Mean accuracy.

Source code in ADLStream/evaluation/metrics.py

def accuracy(y, o):
    """Accuracy score.

    Args:
        y (np.array): Real values.
        o (np.array): Predictions.

    Returns:
        float: Mean accuracy.
    """
    y, o = _preprocess_predictions(y, o)
    return np.mean(y == o)

auc(y, o)

Area under curve (AUC).

TODO Implement AUC

Parameters:

Name	Type	Description	Default
y	np.array	Real values.	required
o	np.array	Predictions.	required

Returns:

Type	Description
float	AUC.

Source code in ADLStream/evaluation/metrics.py

def auc(y, o):
    """Area under curve (AUC).

    TODO Implement AUC

    Args:
        y (np.array): Real values.
        o (np.array): Predictions.

    Returns:
        float: AUC.
    """
    raise NotImplementedError

evaluate(metric, y, o, **kwargs)

Compute a specific loss function given expected output and predicted output.

Parameters:

Name	Type	Description	Default
metric	str	Loss function to use.	required
y	list or np.array	Target data.	required
o	list or np.array	Predictions.	required

Returns:

Type	Description
float	loss

Source code in ADLStream/evaluation/metrics.py

def evaluate(metric, y, o, **kwargs):
    """Compute a specific loss function given expected output and predicted output.

    Args:
        metric (str): Loss function to use.
        y (list or np.array): Target data.
        o (list or np.array): Predictions.

    Returns:
        float: loss
    """
    assert metric.upper() in METRICS, "Metric {} not supported.".format(metric)
    o = np.asarray(o) if type(o) == type([]) else o
    y = np.asarray(y).reshape(o.shape) if type(y) == type([]) else y
    return METRICS[metric.upper()](y, o, **kwargs)

kappa(y, o)

Cohen kappa score.

Parameters:

Name	Type	Description	Default
y	np.array	Real values.	required
o	np.array	Predictions.	required

Returns:

Type	Description
float	Kappa.

Source code in ADLStream/evaluation/metrics.py

def kappa(y, o):
    """Cohen kappa score.

    Args:
        y (np.array): Real values.
        o (np.array): Predictions.

    Returns:
        float: Kappa.
    """
    confusion = _confusion_matrix(y, o)
    n_classes = confusion.shape[0]
    if n_classes < 2:
        return 1
    sum0 = np.sum(confusion, axis=0)
    sum1 = np.sum(confusion, axis=1)
    expected = np.outer(sum0, sum1) / np.sum(sum0)

    w_mat = np.ones([n_classes, n_classes], dtype=np.int)
    w_mat.flat[:: n_classes + 1] = 0

    k = np.sum(w_mat * confusion) / np.sum(w_mat * expected)
    return 1 - k

mae(y, o)

Mean absolute error (MAE).

Parameters:

Name	Type	Description	Default
y	np.array	Real values.	required
o	np.array	Predictions.	required

Returns:

Type	Description
float	MAE.

Source code in ADLStream/evaluation/metrics.py

def mae(y, o):
    """Mean absolute error (MAE).

    Args:
        y (np.array): Real values.
        o (np.array): Predictions.

    Returns:
        float: MAE.
    """
    return np.mean(np.abs(o - y))

mape(y, o)

Mean Absolute Percentage Error (MAPE).

TODO Implement MAPE

Parameters:

Name	Type	Description	Default
y	np.array	Real values.	required
o	np.array	Predictions.	required

Returns:

Type	Description
float	MAPE.

Source code in ADLStream/evaluation/metrics.py

def mape(y, o):
    """Mean Absolute Percentage Error (MAPE).

    TODO Implement MAPE

    Args:
        y (np.array): Real values.
        o (np.array): Predictions.

    Returns:
        float: MAPE.
    """
    raise NotImplementedError

wape(y, o)

Weighted Absolute Percentage Error (WAPE).

TODO Implement WAPE

Parameters:

Name	Type	Description	Default
y	np.array	Real values.	required
o	np.array	Predictions.	required

Returns:

Type	Description
float	WAPE.

Source code in ADLStream/evaluation/metrics.py

def wape(y, o):
    """Weighted Absolute Percentage Error (WAPE).

    TODO Implement WAPE

    Args:
        y (np.array): Real values.
        o (np.array): Predictions.

    Returns:
        float: WAPE.
    """
    raise NotImplementedError