def test_to_prob(an_array, a_binary_array):
out = to_prob(an_array)
assert not np.allclose(out, an_array)
out = to_prob(a_binary_array)
assert not np.allclose(out, a_binary_array)
a_array_scaled = softmax(an_array, 1)
a_binary_array_scaled = expit(a_binary_array)
out = to_prob(a_array_scaled)
assert np.allclose(out, a_array_scaled)
out = to_prob(a_binary_array_scaled)
assert np.allclose(out, a_binary_array_scaled)
def update(self, output=None, target=None):
"""
Update the confusion matrice according to output and target.
Args:
output (tensor): predictions of model
target (tensor): labels
"""
output = output.detach().cpu().numpy()
target = target.detach().cpu().numpy()
output = to_prob(output)
assert output.ndim > target.ndim, "Only multiclass classification is supported."
for cls in range(self.num_classes):
target_cls = (target == cls).astype(np.int8)
for th in self.threshold:
report = self._make_report(output[:, cls, ...], target_cls, th)
self._data[cls][th].fp += report.fp
self._data[cls][th].tp += report.tp
self._data[cls][th].fn += report.fn
def update(self, output=None, target=None):
"""
Updating the true positive (tp) and number of samples in each bin.
Args:
output (tensor): logits or predictions of model
target (tensor): labels
"""
output = output.detach().cpu().numpy()
target = target.detach().cpu().numpy()
output = to_prob(output)
# this is to make sure handling 1.0 value confidence to be assigned to a bin
output = np.clip(output, 0, 0.9999)
for cls in range(self.n_cls):
for pred, t in zip(output[:, cls], target):
bin_id = int(math.floor(pred * self.n_bins))
self.samples[cls, bin_id] += 1
self.tp[cls, bin_id] += int(cls == t)
def update(self, output=None, target=None):
"""
Updating the true positive (tp) and number of samples in each bin.
Args:
output (tensor): logits or predictions of model
target (tensor): labels
"""
output = output.detach().cpu().numpy()
target = target.detach().cpu().numpy()
output = to_prob(output)
# this is to make sure handling 1.0 value confidence to be assigned to a bin
output = np.clip(output, 0, 0.9999)
for pred, t in zip(output, target):
t = int(t) # Force the conversion
conf, p_cls = pred.max(), pred.argmax()
bin_id = int(math.floor(conf * self.n_bins))
self.samples[t, bin_id] += 1
self.tp[t, bin_id] += int(p_cls == t)
def wrapper(self, probabilities):
# Expected shape : [n_sample, n_classes, ..., n_iterations]
probabilities = to_prob(probabilities)
return fn(self, probabilities)
