def update(self, input, target):
self.y_true = tensor_to_numpy(target)
if self.normalizate and self.task_type == 'binary':
y_prob = tensor_to_numpy(input.sigmoid().data)
elif self.normalizate and self.task_type == 'multiclass':
y_prob = tensor_to_numpy(input.softmax(-1).data)
else:
y_prob = tensor_to_numpy(input)
if self.task_type == 'binary':
if self.thresh and self.search_thresh == False:
self.y_pred = (y_prob > self.thresh).astype(int)
self.value()
else:
thresh, f1 = self.thresh_search(y_prob=y_prob)
print(f"Best thresh: {thresh:.4f} - F1 Score: {f1:.4f}")
if self.task_type == 'multiclass':
self.y_pred = np.argmax(y_prob, 1)
def update(self, input, target):
if self.task_type == 'binary':
self.y_prob = tensor_to_numpy(input.sigmoid().data)
else:
self.y_prob = tensor_to_numpy(input.softmax(-1).data)
self.y_true = tensor_to_numpy(target)
def value(self):
return matthews_corrcoef(tensor_to_numpy(self.labels),
tensor_to_numpy(self.preds))
def update(self, input, target):
self.y_prob = tensor_to_numpy(input.sigmoid().data)
self.y_true = tensor_to_numpy(target)
def update(self, input, target):
_, y_pred = torch.max(input, 1)
self.y_pred = tensor_to_numpy(y_pred)
self.y_true = tensor_to_numpy(target)