def test_AETrainer():
"""Test the AETrainer class
"""
net = MultiDAE_net([1, 2], [2, 1], .1)
model = AETrainer(net)
assert hasattr(model, "network"), "model should have the attribute newtork"
assert hasattr(model, "device"), "model should have the attribute device"
assert hasattr(
model,
"learning_rate"), "model should have the attribute learning_rate"
assert hasattr(model,
"optimizer"), "model should have the attribute optimizer"
assert model.learning_rate == 1e-3, "the learning rate should be 1e-3"
assert model.network == net, "the network should be the same as the parameter"
assert model.device == torch.device("cpu"), "the device should be cpu"
assert isinstance(model.optimizer,
torch.optim.Adam), "optimizer should be of Adam type"
assert str(model) == repr(
model), "repr and str should have the same effect"
gt = torch.FloatTensor([[1, 1], [2, 1]])
pred = torch.FloatTensor([[1, 1], [1, 1]])
assert model.loss_function(pred, gt) == torch.FloatTensor(
[.25]), "the loss should be .25"
values = np.array([1., 1., 1.])
rows = np.array([0, 0, 1])
cols = np.array([0, 1, 1])
train = csr_matrix((values, (rows, cols)))
sampler = DataSampler(train, batch_size=1, shuffle=False)
x = torch.FloatTensor([[1, 1], [2, 2]])
model.predict(x, True)
torch.manual_seed(12345)
out_1 = model.predict(x, False)[0]
model.train(sampler, num_epochs=10, verbose=4)
torch.manual_seed(12345)
out_2 = model.predict(x, False)[0]
assert not torch.all(out_1.eq(out_2)), "the outputs should be different"
tmp = tempfile.NamedTemporaryFile()
model.save_model(tmp.name, 1)
net = MultiDAE_net([1, 2], [2, 1], .1)
model2 = AETrainer(net)
model2.load_model(tmp.name)
torch.manual_seed(12345)
out_1 = model.predict(x, False)[0]
torch.manual_seed(12345)
out_2 = model2.predict(x, False)[0]
assert torch.all(out_1.eq(out_2)), "the outputs should be the same"
sampler = DataSampler(train, train, batch_size=1, shuffle=False)
res = model.validate(sampler, "ndcg@1")
assert isinstance(res, np.ndarray), "results should the be a numpy array"
assert len(res) == 2, "results should be of length 2"
def test_TorchNNTrainer():
"""Test the TorchNNTrainer class
"""
net = MultiDAE_net([1, 2], [2, 1], .1)
model = TorchNNTrainer(net)
assert hasattr(model, "network"), "model should have the attribute newtork"
assert hasattr(model, "device"), "model should have the attribute device"
assert hasattr(
model,
"learning_rate"), "model should have the attribute learning_rate"
assert hasattr(model,
"optimizer"), "model should have the attribute optimizer"
assert model.learning_rate == 1e-3, "the learning rate should be 1e-3"
assert model.network == net, "the network should be the same as the parameter"
assert model.device == torch.device("cpu"), "the device should be cpu"
assert model.optimizer is None, "optimizer should be None"
assert str(model) == repr(model)
x = torch.FloatTensor([[1, 1], [2, 2]])
with pytest.raises(NotImplementedError):
model.loss_function(None, None)
model.train(None, None)
model.train_epoch(0, None)
model.train_batch(0, None, None)
model.predict(x)
def test_MultiDAE_net():
"""Test the MultiDAE_net class
"""
net = MultiDAE_net([1, 2], [2, 1], .1)
x = torch.FloatTensor([[1, 1], [2, 2]])
y = net(x)
assert hasattr(net, "enc_dims"), "Missing enc_dims attribute"
assert hasattr(net, "dec_dims"), "Missing dec_dims attribute"
assert hasattr(net, "dropout"), "Missing dropout attribute"
assert hasattr(net, "dec_layers"), "Missing dec_layers attribute"
assert hasattr(net, "enc_layers"), "Missing end_layers attribute"
assert isinstance(net.dropout, torch.nn.Dropout), "dropout must be a torch.nn.Dropout"
assert net.dropout.p == .1, "dropout probability must be equal to .1"
assert isinstance(y, torch.FloatTensor), "y should be a torch.FloatTensor"
assert y.shape == x.shape, "The shape of x and y should be the same"