class WindowMultiTargetClassificationMeasurements(BaseObject):
""" This class will maintain a fixed sized window of the newest information
about one classifier. It can provide, as requested, any of the relevant
current metrics about the classifier, measured inside the window.
This class will keep updated statistics about a multi output classifier,
using a confusion matrix adapted to multi output problems, the
MOLConfusionMatrix, alongside other of the classifier's relevant
attributes stored in ComplexFastBuffer objects, which will simulate
fixed sized windows.
Its functionality is somewhat similar to those of the
MultiTargetClassificationMeasurements class. The difference is that the statistics
kept by this class are local, or partial, while the statistics kept by
the MultiTargetClassificationMeasurements class are global.
At any given moment, it can compute the following statistics: hamming_loss,
hamming_score, exact_match and j_index.
Parameters
----------
targets: list
A list containing the possible labels.
dtype: data type (Default: numpy.int64)
The data type of the existing labels.
window_size: int (Default: 200)
The width of the window. Determines how many samples the object
can see.
Examples
--------
"""
def __init__(self, targets=None, dtype=np.int64, window_size=200):
super().__init__()
if targets is not None:
self.n_targets = len(targets)
else:
self.n_targets = 0
self.confusion_matrix = MOLConfusionMatrix(self.n_targets, dtype)
self.last_true_label = None
self.last_prediction = None
self.targets = targets
self.window_size = window_size
self.exact_match_count = 0
self.j_sum = 0
self.true_labels = FastComplexBuffer(window_size, self.n_targets)
self.predictions = FastComplexBuffer(window_size, self.n_targets)
def reset(self):
if self.targets is not None:
self.n_targets = len(self.targets)
else:
self.n_targets = 0
self.confusion_matrix.restart(self.n_targets)
self.last_true_label = None
self.last_prediction = None
self.exact_match_count = 0
self.j_sum = 0
self.true_labels = FastComplexBuffer(self.window_size, self.n_targets)
self.predictions = FastComplexBuffer(self.window_size, self.n_targets)
def add_result(self, y_true, y_pred):
""" Updates its statistics with the results of a prediction.
Adds the result to the MOLConfusionMatrix, and updates the
ComplexFastBuffer objects.
Parameters
----------
y_true: list or numpy.ndarray
The true label.
y_pred: list or numpy.ndarray
The classifier's prediction
"""
self.last_true_label = y_true
self.last_prediction = y_pred
m = 0
if hasattr(y_true, 'size'):
m = y_true.size
elif hasattr(y_true, 'append'):
m = len(y_true)
self.n_targets = m
for i in range(m):
self.confusion_matrix.update(i, y_true[i], y_pred[i])
old_true = self.true_labels.add_element(y_true)
old_predict = self.predictions.add_element(y_pred)
if (old_true is not None) and (old_predict is not None):
for i in range(m):
self.confusion_matrix.remove(old_true[0][i], old_predict[0][i])
def get_last(self):
return self.last_true_label, self.last_prediction
def get_hamming_loss(self):
""" Computes the window/current Hamming loss, which is the
complement of the Hamming score metric.
Returns
-------
float
The window/current hamming loss.
"""
return 1.0 - self.get_hamming_score()
def get_hamming_score(self):
""" Computes the window/current Hamming score, defined as the number of
correctly classified labels divided by the total number of labels
classified.
Returns
-------
float
The window/current hamming score.
"""
return hamming_score(self.true_labels.get_queue(),
self.predictions.get_queue())
def get_exact_match(self):
""" Computes the window/current exact match metric.
This is the most strict multi output metric, defined as the number of
samples that have all their labels correctly classified, divided by the
total number of samples.
Returns
-------
float
The window/current exact match metric.
"""
return exact_match(self.true_labels.get_queue(),
self.predictions.get_queue())
def get_j_index(self):
""" Computes the window/current Jaccard index, also known as the intersection
over union metric. It is calculated by dividing the number of correctly
classified labels by the union of predicted and true labels.
Returns
-------
float
The window/current Jaccard index.
"""
return j_index(self.true_labels.get_queue(),
self.predictions.get_queue())
def get_total_sum(self):
return self.confusion_matrix.get_total_sum()
@property
def matrix(self):
return self.confusion_matrix.matrix
@property
def sample_count(self):
return self.true_labels.get_current_size()
def get_info(self):
return '{}:'.format(type(self).__name__) + \
' - sample_count: {}'.format(self.sample_count) + \
' - hamming_loss: {:.6f}'.format(self.get_hamming_loss()) + \
' - hamming_score: {:.6f}'.format(self.get_hamming_score()) + \
' - exact_match: {:.6f}'.format(self.get_exact_match()) + \
' - j_index: {:.6f}'.format(self.get_j_index())
def get_class_type(self):
return 'measurement'
class MultiTargetClassificationMeasurements(BaseObject):
""" This class will keep updated statistics about a multi output classifier,
using a confusion matrix adapted to multi output problems, the
MOLConfusionMatrix, alongside other relevant attributes.
The performance metrics for multi output tasks are different from those used
for normal classification tasks. Thus, the statistics provided by this class
are different from those provided by the ClassificationMeasurements and from
the WindowClassificationMeasurements.
At any given moment, it can compute the following statistics: hamming_loss,
hamming_score, exact_match and j_index.
Parameters
----------
targets: list
A list containing the possible labels.
dtype: data type (Default: numpy.int64)
The data type of the existing labels.
Examples
--------
"""
def __init__(self, targets=None, dtype=np.int64):
super().__init__()
if targets is not None:
self.n_targets = len(targets)
else:
self.n_targets = 0
self.confusion_matrix = MOLConfusionMatrix(self.n_targets, dtype)
self.last_true_label = None
self.last_prediction = None
self.sample_count = 0
self.targets = targets
self.exact_match_count = 0
self.j_sum = 0
def reset(self):
if self.targets is not None:
self.n_targets = len(self.targets)
else:
self.n_targets = 0
self.confusion_matrix.restart(self.n_targets)
self.last_true_label = None
self.last_prediction = None
self.sample_count = 0
self.exact_match_count = 0
self.j_sum = 0
def add_result(self, y_true, y_pred):
""" Updates its statistics with the results of a prediction.
Adds the result to the MOLConfusionMatrix and update exact_matches and
j-index sum counts.
Parameters
----------
y_true: list or numpy.ndarray
The true label.
y_pred: list or numpy.ndarray
The classifier's prediction
"""
self.last_true_label = y_true
self.last_prediction = y_pred
m = 0
if isinstance(y_true, np.ndarray):
m = y_true.size
elif isinstance(y_true, list):
m = len(y_true)
self.n_targets = m
equal = True
for i in range(m):
self.confusion_matrix.update(i, y_true[i], y_pred[i])
# update exact_match count
if y_true[i] != y_pred[i]:
equal = False
# update exact_match
if equal:
self.exact_match_count += 1
# update j_index count
inter = sum((y_true * y_pred) > 0) * 1.
union = sum((y_true + y_pred) > 0) * 1.
if union > 0:
self.j_sum += inter / union
elif np.sum(y_true) == 0:
self.j_sum += 1
self.sample_count += 1
def get_last(self):
return self.last_true_label, self.last_prediction
def get_hamming_loss(self):
""" Computes the Hamming loss, which is the complement of the
Hamming score metric.
Returns
-------
float
The hamming loss.
"""
return 1.0 - self.get_hamming_score()
def get_hamming_score(self):
""" Computes the Hamming score, defined as the number of correctly
classified labels divided by the total number of labels classified.
Returns
-------
float
The Hamming score.
"""
try:
return self.confusion_matrix.get_sum_main_diagonal() / (
self.sample_count * self.n_targets)
except ZeroDivisionError:
return 0.0
def get_exact_match(self):
""" Computes the exact match metric.
This is the most strict multi output metric, defined as the number of
samples that have all their labels correctly classified, divided by the
total number of samples.
Returns
-------
float
The exact match metric.
"""
return self.exact_match_count / self.sample_count
def get_j_index(self):
""" Computes the Jaccard index, also known as the intersection over union
metric. It is calculated by dividing the number of correctly classified
labels by the union of predicted and true labels.
Returns
-------
float
The Jaccard index.
"""
return self.j_sum / self.sample_count
def get_total_sum(self):
return self.confusion_matrix.get_total_sum()
@property
def _matrix(self):
return self.confusion_matrix.matrix
def get_info(self):
return '{}:'.format(type(self).__name__) + \
' - sample_count: {}'.format(self.sample_count) + \
' - hamming_loss: {:.6f}'.format(self.get_hamming_loss()) + \
' - hamming_score: {:.6f}'.format(self.get_hamming_score()) + \
' - exact_match: {:.6f}'.format(self.get_exact_match()) + \
' - j_index: {:.6f}'.format(self.get_j_index())
def get_class_type(self):
return 'measurement'
