def test_learner_w_multilabel_string_and_int_label_datasets_ensemble(
mixed_multi_label_ensemble, task_dicts_string_labels,
multi_label_loss_acc_dicts):
train_dataset_loader, val_dataset_loader = mixed_multi_label_ensemble
task_dict, image_task_dict = task_dicts_string_labels
loss_dict, acc_dict = multi_label_loss_acc_dicts
model = BertResnetEnsembleForMultiTaskClassification(
image_task_dict=image_task_dict)
learn = MultiInputMultiTaskLearner(model,
train_dataset_loader,
val_dataset_loader,
task_dict,
loss_function_dict=loss_dict,
metric_function_dict=acc_dict)
assert learn.train_dataloader.label_mappings['task1'] == {
0: 'cat',
1: 'dog'
}
assert learn.train_dataloader.label_mappings['task2'] == {0: 1, 1: 2}
assert learn.val_dataloader.label_mappings['task1'] == {0: 'cat', 1: 'dog'}
assert learn.val_dataloader.label_mappings['task2'] == {0: 1, 1: 2}
def test_learner_with_missing_loss(test_missing_task, test_train_val_loaders):
loss_function_dict = test_missing_task
task_dict = {'task1': 1, 'task2': 1}
model = ResnetForMultiTaskClassification(new_task_dict=task_dict)
with pytest.raises(Exception) as e:
MultiInputMultiTaskLearner(model, test_train_val_loaders,
test_train_val_loaders, task_dict,
loss_function_dict)
assert ('make sure all tasks are contained in the' in str(e.value)) is True
def test_learner_with_invalid_loss_name(test_invalid_loss_name,
test_train_val_loaders):
loss_function_dict = test_invalid_loss_name
task_dict = {'task1': 1, 'task2': 1}
model = ResnetForMultiTaskClassification(new_task_dict=task_dict)
with pytest.raises(Exception) as e:
MultiInputMultiTaskLearner(model, test_train_val_loaders,
test_train_val_loaders, task_dict,
loss_function_dict)
assert ('Found invalid loss function:' in str(e.value)) is True
def test_learner_with_none_loss_dict(test_no_loss_dict,
test_train_val_loaders):
loss_function_dict = test_no_loss_dict
task_dict = {'task1': 1, 'task2': 1}
model = ResnetForMultiTaskClassification(new_task_dict=task_dict)
learn = MultiInputMultiTaskLearner(model, test_train_val_loaders,
test_train_val_loaders, task_dict,
loss_function_dict)
assert str(learn.loss_function_dict['task1']) == 'CrossEntropyLoss()'
assert str(learn.loss_function_dict['task2']) == 'CrossEntropyLoss()'
def test_learner_with_bce_and_cce(test_all_tasks_diff_losses,
test_train_val_loaders):
loss_function_dict = test_all_tasks_diff_losses
task_dict = {'task1': 1, 'task2': 1}
model = ResnetForMultiTaskClassification(new_task_dict=task_dict)
learn = MultiInputMultiTaskLearner(model, test_train_val_loaders,
test_train_val_loaders, task_dict,
loss_function_dict)
assert str(learn.loss_function_dict['task1']) == 'BCEWithLogitsLoss()'
assert str(learn.loss_function_dict['task2']) == 'CrossEntropyLoss()'
def test_learner_with_none_acc_dict(test_no_acc_dict, test_train_val_loaders):
acc_function_dict = test_no_acc_dict
task_dict = {'task1': 1, 'task2': 1}
model = ResnetForMultiTaskClassification(new_task_dict=task_dict)
learn = MultiInputMultiTaskLearner(model,
test_train_val_loaders,
test_train_val_loaders,
task_dict,
metric_function_dict=acc_function_dict)
assert 'multi_class_accuracy' in str(learn.metric_function_dict['task1'])
assert 'multi_class_accuracy' in str(learn.metric_function_dict['task2'])
def test_learner_w_string_datasets_not_matching_categories_image(
mixed_labels, task_dicts_string_labels):
train_dataset_loader, val_dataset_loader = mixed_labels
val_dataset_loader.label_mappings['task1'] = {0: 'blue', 3: 'green'}
task_dict, _ = task_dicts_string_labels
model = ResnetForMultiTaskClassification(new_task_dict=task_dict)
with pytest.raises(ValueError) as e:
MultiInputMultiTaskLearner(model, train_dataset_loader,
val_dataset_loader, task_dict)
assert ('Mapping mismatch in task1 task. Check that all categories' in str(
e.value)) is True
def test_learner_w_string_datasets_not_matching_categoies_text(
mixed_labels_text, task_dicts_string_labels):
train_dataset_loader, val_dataset_loader = mixed_labels_text
val_dataset_loader.label_mappings['task1'] = {0: 'blue', 3: 'green'}
task_dict, _ = task_dicts_string_labels
model = BertForMultiTaskClassification.from_pretrained(
'bert-base-uncased', new_task_dict=task_dict)
with pytest.raises(ValueError) as e:
MultiInputMultiTaskLearner(model, train_dataset_loader,
val_dataset_loader, task_dict)
assert ('Mapping mismatch in task1 task. Check that all categories' in str(
e.value)) is True
def test_learner_w_string_and_int_label_datasets_success_ensemble(
mixed_labels_ensemble, task_dicts_string_labels):
train_dataset_loader, val_dataset_loader = mixed_labels_ensemble
task_dict, image_task_dict = task_dicts_string_labels
model = BertResnetEnsembleForMultiTaskClassification(
image_task_dict=image_task_dict)
learn = MultiInputMultiTaskLearner(model, train_dataset_loader,
val_dataset_loader, task_dict)
assert learn.train_dataloader.label_mappings['task1'] == {
0: 'cat',
1: 'dog'
}
assert learn.train_dataloader.label_mappings['task2'] == {0: 1, 1: 2}
assert learn.val_dataloader.label_mappings['task1'] == {0: 'cat', 1: 'dog'}
assert learn.val_dataloader.label_mappings['task2'] == {0: 1, 1: 2}
def test_w_string_multilabel_not_matching_categories_multilabel_text(
multi_label_mixed_labels_text, task_dicts_string_labels,
multi_label_loss_acc_dicts):
train_dataset_loader, val_dataset_loader = multi_label_mixed_labels_text
val_dataset_loader.label_mappings['task1'] = {0: 'blue', 3: 'green'}
task_dict, _ = task_dicts_string_labels
loss_dict, acc_dict = multi_label_loss_acc_dicts
model = ResnetForMultiTaskClassification(new_task_dict=task_dict)
with pytest.raises(ValueError) as e:
MultiInputMultiTaskLearner(model,
train_dataset_loader,
val_dataset_loader,
task_dict,
loss_function_dict=loss_dict,
metric_function_dict=acc_dict)
assert ('Mapping mismatch in task1 task. Check that all categories' in str(
e.value)) is True
def test_learner_w_string_and_int_label_datasets_success_text(
mixed_labels_text, task_dicts_string_labels):
train_dataset_loader, val_dataset_loader = mixed_labels_text
task_dict, _ = task_dicts_string_labels
model = BertForMultiTaskClassification.from_pretrained(
'bert-base-uncased', new_task_dict=task_dict)
learn = MultiInputMultiTaskLearner(model, train_dataset_loader,
val_dataset_loader, task_dict)
assert learn.train_dataloader.label_mappings['task1'] == {
0: 'cat',
1: 'dog'
}
assert learn.train_dataloader.label_mappings['task2'] == {0: 1, 1: 2}
assert learn.val_dataloader.label_mappings['task1'] == {0: 'cat', 1: 'dog'}
assert learn.val_dataloader.label_mappings['task2'] == {0: 1, 1: 2}
def test_learner_w_multilabel_string_and_int_label_datasets_text(
multi_label_mixed_labels_text, task_dicts_string_labels,
multi_label_loss_acc_dicts):
train_dataset_loader, val_dataset_loader = multi_label_mixed_labels_text
task_dict, _ = task_dicts_string_labels
loss_dict, acc_dict = multi_label_loss_acc_dicts
model = ResnetForMultiTaskClassification(new_task_dict=task_dict)
learn = MultiInputMultiTaskLearner(model,
train_dataset_loader,
val_dataset_loader,
task_dict,
loss_function_dict=loss_dict,
metric_function_dict=acc_dict)
assert learn.train_dataloader.label_mappings['task1'] == {
0: 'cat',
1: 'dog'
}
assert learn.train_dataloader.label_mappings['task2'] == {0: 1, 1: 2}
assert learn.val_dataloader.label_mappings['task1'] == {0: 'cat', 1: 'dog'}
assert learn.val_dataloader.label_mappings['task2'] == {0: 1, 1: 2}
