def meta_learn(self, batch, batch_idx, ways, shots, queries):
self.features.train()
learner = self.classifier.clone()
learner.train()
data, labels = batch
data = self.features(data)
# Separate data into adaptation and evaluation sets
support_indices = np.zeros(data.size(0), dtype=bool)
selection = np.arange(ways) * (shots + queries)
for offset in range(shots):
support_indices[selection + offset] = True
query_indices = torch.from_numpy(~support_indices)
support_indices = torch.from_numpy(support_indices)
support = data[support_indices]
support_labels = labels[support_indices]
query = data[query_indices]
query_labels = labels[query_indices]
# Adapt the classifier
for step in range(self.adaptation_steps):
preds = learner(support)
train_error = self.loss(preds, support_labels)
learner.adapt(train_error)
# Evaluating the adapted model
predictions = learner(query)
valid_error = self.loss(predictions, query_labels)
valid_accuracy = accuracy(predictions, query_labels)
return valid_error, valid_accuracy
def meta_learn(self, batch, batch_idx, ways, shots, queries):
self.features.train()
data, labels = batch
# Sort data samples by labels
sort = torch.sort(labels)
data = data.squeeze(0)[sort.indices].squeeze(0)
labels = labels.squeeze(0)[sort.indices].squeeze(0)
# Compute support and query embeddings
embeddings = self.features(data)
support_indices = np.zeros(data.size(0), dtype=bool)
selection = np.arange(ways) * (shots + queries)
for offset in range(shots):
support_indices[selection + offset] = True
query_indices = torch.from_numpy(~support_indices)
support_indices = torch.from_numpy(support_indices)
support = embeddings[support_indices]
support_labels = labels[support_indices]
query = embeddings[query_indices]
query_labels = labels[query_indices]
self.classifier.fit_(support, support_labels)
logits = self.classifier(query)
eval_loss = self.loss(logits, query_labels)
eval_accuracy = accuracy(logits, query_labels)
return eval_loss, eval_accuracy
def test_simple(self):
for normalize in [True, False]:
X = []
y = []
for i in range(NUM_CLASSES):
images = torch.randn(1, *IMAGE_SHAPES).expand(NUM_SHOTS, *IMAGE_SHAPES)
labels = torch.ones(NUM_SHOTS).long()
X.append(images)
y.append(i * labels)
X = torch.cat(X, dim=0)
y = torch.cat(y)
X.requires_grad = True
X_support = X + torch.randn_like(X) * NOISE
X_query = X + torch.randn_like(X) * NOISE
# Compute embeddings
X_support = X_support.view(NUM_CLASSES * NUM_SHOTS, -1)
X_query = X_query.view(NUM_CLASSES * NUM_SHOTS, -1)
classifier = l2l.nn.SVClassifier(
support=X_support,
labels=y,
normalize=normalize,
)
predictions = classifier(X_query)
acc = accuracy(predictions, y)
self.assertTrue(acc >= 0.95)
def test_simple(self):
for distance in ["euclidean", "cosine"]:
for normalize in [True, False]:
# Create some fake data
X = []
y = []
for i in range(NUM_CLASSES):
images = torch.randn(1, *IMAGE_SHAPES).expand(
NUM_SHOTS, *IMAGE_SHAPES)
labels = torch.ones(NUM_SHOTS).long()
X.append(images)
y.append(i * labels)
X = torch.cat(X, dim=0)
y = torch.cat(y)
X_support = X + torch.randn_like(X) * NOISE
X_query = X + torch.randn_like(X) * NOISE
# Compute embeddings
X_support = X_support.view(NUM_CLASSES * NUM_SHOTS, -1)
X_query = X_query.view(NUM_CLASSES * NUM_SHOTS, -1)
classifier = l2l.nn.PrototypicalClassifier(
support=X_support,
labels=y,
distance=distance,
normalize=normalize,
)
predictions = classifier(X_query)
acc = accuracy(predictions, y)
self.assertTrue(acc >= 0.95)
for normalize in [True, False]:
X = []
y = []
for i in range(NUM_CLASSES):
images = torch.randn(1,
*IMAGE_SHAPES).expand(NUM_SHOTS,
*IMAGE_SHAPES)
labels = torch.ones(NUM_SHOTS).long()
X.append(images)
y.append(i * labels)
X = torch.cat(X, dim=0)
y = torch.cat(y)
X.requires_grad = True
X = X.cuda()
y = y.cuda()
X_support = X + torch.randn_like(X) * NOISE
X_query = X + torch.randn_like(X) * NOISE
# Compute embeddings
X_support = X_support.view(NUM_CLASSES * NUM_SHOTS, -1)
X_query = X_query.view(NUM_CLASSES * NUM_SHOTS, -1)
classifier = SVClassifier(
support=X_support,
labels=y,
normalize=normalize,
)
predictions = classifier(X_query)
acc = accuracy(predictions, y)
assert acc >= 0.95
