def test_construction_model(self):
mlp = MLP(hidden_layer_sizes=(2, ))
# number of data = 3
# number of feature = 2
x_train = np.zeros([3, 2])
y_train = [1, 0, 1]
mlp.train(x_train, y_train)
self.assertTrue(mlp.input.output_list[0].output_list[0].output_list[0].
name.find('softmax') < 0)
def train(args):
perturb_mock, sgRNA_list_mock = makedata.json_to_perturb_data(path = "/home/member/xywang/WORKSPACE/MaryGUO/one-shot/MOCK_MON_crispr_combine/crispr_analysis")
total = sc.read_h5ad("/home/member/xywang/WORKSPACE/MaryGUO/one-shot/mock_one_perturbed.h5ad")
trainset, testset = preprocessing.make_total_data(total,sgRNA_list_mock)
TrainSet = perturbdataloader(trainset, ways = args.num_ways, support_shots = args.num_shots, query_shots = 15)
TrainLoader = DataLoader(TrainSet, batch_size=args.batch_size_train, shuffle=False,num_workers=args.num_workers)
model = MLP(out_features = args.num_ways)
model.to(device=args.device)
model.train()
meta_optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)
# Training loop
with tqdm(TrainLoader, total=args.num_batches) as pbar:
for batch_idx, (inputs_support, inputs_query, target_support, target_query) in enumerate(pbar):
model.zero_grad()
inputs_support = inputs_support.to(device=args.device)
target_support = target_support.to(device=args.device)
inputs_query = inputs_query.to(device=args.device)
target_query = target_query.to(device=args.device)
outer_loss = torch.tensor(0., device=args.device)
accuracy = torch.tensor(0., device=args.device)
for task_idx, (train_input, train_target, test_input,
test_target) in enumerate(zip(inputs_support, target_support,inputs_query, target_query)):
train_logit = model(train_input)
inner_loss = F.cross_entropy(train_logit, train_target)
model.zero_grad()
params = gradient_update_parameters(model,
inner_loss,
step_size=args.step_size,
first_order=args.first_order)
test_logit = model(test_input, params=params)
outer_loss += F.cross_entropy(test_logit, test_target)
with torch.no_grad():
accuracy += get_accuracy(test_logit, test_target)
outer_loss.div_(args.batch_size_train)
accuracy.div_(args.batch_size_train)
outer_loss.backward()
meta_optimizer.step()
pbar.set_postfix(accuracy='{0:.4f}'.format(accuracy.item()))
if batch_idx >= args.num_batches or accuracy.item() > 0.95:
break
# Save model
if args.output_folder is not None:
filename = os.path.join(args.output_folder, 'maml_omniglot_'
'{0}shot_{1}way.th'.format(args.num_shots, args.num_ways))
with open(filename, 'wb') as f:
state_dict = model.state_dict()
torch.save(state_dict, f)
# start test
test_support, test_query, test_target_support, test_target_query \
= helpfuntions.sample_once(testset,support_shot=args.num_shots, shuffle=False,plus = len(trainset))
test_query = torch.from_numpy(test_query).to(device=args.device)
test_target_query = torch.from_numpy(test_target_query).to(device=args.device)
TrainSet = perturbdataloader_test(test_support, test_target_support)
TrainLoader = DataLoader(TrainSet, args.batch_size_test)
meta_optimizer.zero_grad()
inner_losses = []
accuracy_test = []
for epoch in range(args.num_epoch):
model.to(device=args.device)
model.train()
for _, (inputs_support,target_support) in enumerate(TrainLoader):
inputs_support = inputs_support.to(device=args.device)
target_support = target_support.to(device=args.device)
train_logit = model(inputs_support)
loss = F.cross_entropy(train_logit, target_support)
inner_losses.append(loss)
loss.backward()
meta_optimizer.step()
meta_optimizer.zero_grad()
test_logit = model(test_query)
with torch.no_grad():
accuracy = get_accuracy(test_logit, test_target_query)
accuracy_test.append(accuracy)
if (epoch + 1) % 3 == 0:
print('Epoch [{}/{}], Loss: {:.4f},accuray: {:.4f}'.format(epoch + 1, args.num_epoch, loss,accuracy))
