def test_initialisation(self):
module = SimpleMLP()
module = module.to(self.device)
old_classifier_weight = torch.clone(module.classifier.weight)
old_classifier_bias = torch.clone(module.classifier.bias)
module = as_multitask(module, "classifier")
module = module.to(self.device)
new_classifier_weight = torch.clone(
module.classifier.classifiers["0"].classifier.weight)
new_classifier_bias = torch.clone(
module.classifier.classifiers["0"].classifier.bias)
self.assertTrue(
torch.equal(old_classifier_weight, new_classifier_weight))
self.assertTrue(torch.equal(old_classifier_bias, new_classifier_bias))
def _test_outputs(self, module, clf_name):
test_input = torch.rand(10, 3, 32, 32)
test_input = test_input.to(self.device)
module_singletask = copy.deepcopy(module)
module_multitask = as_multitask(module, clf_name)
module_multitask = module_multitask.to(self.device)
module_singletask = module_singletask.to(self.device)
# Put in eval mode to deactivate dropouts
module_singletask.eval()
module_multitask.eval()
out_single_task = module_singletask(test_input)
out_multi_task = module_multitask(test_input, task_labels=0)
self.assertTrue(torch.equal(out_single_task, out_multi_task))
def _test_integration(self, module, clf_name, plugins=[]):
module = as_multitask(module, clf_name)
module = module.to(self.device)
optimizer = SGD(
module.parameters(), lr=0.05, momentum=0.9, weight_decay=0.0002
)
strategy = Naive(
module,
optimizer,
train_mb_size=32,
eval_mb_size=32,
device=self.device,
plugins=plugins,
)
for t, experience in enumerate(self.benchmark.train_stream):
strategy.train(experience)
strategy.eval(self.benchmark.test_stream[: t + 1])
def _test_modules(self, module, clf_name):
old_param_total = sum([torch.numel(p) for p in module.parameters()])
module = as_multitask(module, clf_name)
module = module.to(self.device)
self.assertIsInstance(module, MultiTaskModule)
self.assertIsInstance(getattr(module, clf_name), MultiHeadClassifier)
test_input = torch.ones(5, 3, 32, 32)
task_labels = torch.zeros(5, dtype=torch.long)
test_input = test_input.to(self.device)
task_labels = task_labels.to(self.device)
# One task label
output = module(test_input, task_labels=0)
# Several ones
output = module(test_input, task_labels=task_labels)
# Change attribute
module.non_module_attribute = 10
self.assertEqual(module.model.non_module_attribute, 10)
module.non_module_attribute += 5
# Extract params and state dict
new_param_total = sum([torch.numel(p) for p in module.parameters()])
self.assertEqual(new_param_total, old_param_total)
state = module.state_dict()
# Functions returning references
module = module.train()
self.assertIsInstance(module, MultiTaskModule)
self.assertTrue(module.training)
module = module.eval()
self.assertIsInstance(module, MultiTaskModule)
self.assertFalse(module.training)
if self.device == "cuda":
module = module.cuda()
self.assertIsInstance(module, MultiTaskModule)