| import torch.nn as nn |
|
|
|
|
| class MultiTaskModel(nn.Module): |
| def __init__(self, base_net, task_heads, device): |
| super().__init__() |
| self.base_net = base_net |
| self.task_heads = nn.ModuleList(task_heads) |
| self.device = device |
|
|
| def forward(self, x): |
|
|
| base_output = self.base_net(x) |
| |
| task_outputs = [head(base_output) for head in self.task_heads] |
|
|
| return task_outputs |
|
|
| def predict(self, x): |
|
|
| base_output = self.base_net(x) |
| |
| task_outputs = [head.predict(base_output) for head in self.task_heads] |
|
|
| return task_outputs |
|
|
| def accuracy(self, predictions, targets): |
| accuracies = [head.accuracy(prediction, target) for head, prediction, target in zip( |
| self.task_heads, predictions, targets)] |
| return accuracies |
|
|
| def recall(self, predictions, targets): |
| recalls = [head.recall(prediction, target) for head, prediction, target in zip( |
| self.task_heads, predictions, targets)] |
| return recalls |
| |
| def precision(self, predictions, targets): |
| precisions = [head.precision(prediction, target) for head, prediction, target in zip( |
| self.task_heads, predictions, targets)] |
| return precisions |
