def compute_average_precision(true_labels, predictions, verbose=0):
average_precisions = []
for ii in sorted(true_labels.keys()):
ap = average_precision(true_labels[ii], predictions[ii])
average_precisions.append(ap)
if verbose:
print "%3d %6.2f" % (ii, 100 * ap)
if verbose:
print '----------'
print "mAP %6.2f" % (100 * np.mean(average_precisions))
def compute_average_precision(true_labels, predictions, verbose=0):
average_precisions = []
for ii in sorted(true_labels.keys()):
ap = average_precision(true_labels[ii], predictions[ii])
average_precisions.append(ap)
if verbose:
print "%3d %6.2f" % (ii, 100 * ap)
if verbose:
print '----------'
print "mAP %6.2f" % (100 * np.mean(average_precisions))
def fit_late_fusion(self, scores, tr_labels):
# Equal weights.
#D = scores.shape[1]
#self.weights = np.array([1. / D] * D)
# Dumb crossvalidation.
best_ap = 0
D = scores.shape[1]
for self.weights in weights_grid(D):
ap = average_precision(
tr_labels, self.predict_late_fusion(scores))
if self.weights in self.weight_scores:
self.weight_scores[self.weights].append(ap)
else:
self.weight_scores[self.weights] = [ap]
def score(self, te_kernel, te_labels):
return average_precision(te_labels, self.predict_proba(te_kernel))
def score(self, te_kernel, te_labels):
predicted = self.predict(te_kernel)
score = average_precision(te_labels, predicted) * 100
return score
