def test_sequential_4():
with pytest.raises(RuntimeError,
match='illegal parameter type of <h_neurons>'):
SequentialLinear(10, 5, h_neurons=('NaN', ))
with pytest.raises(
RuntimeError,
match='number of parameter not consistent with number of layers'):
SequentialLinear(10, 5, h_neurons=(4, ), h_dropouts=(1, 5))
m = SequentialLinear(10, 1, h_neurons=(3, 4))
tmp = (1, 2)
assert id(m._check_input(tmp)) == id(tmp)
def test_persist_save_checkpoints(data):
class _Trainer(BaseRunner):
def __init__(self):
super().__init__()
self.model = SequentialLinear(50, 2)
def predict(self, x_, y_): # noqa
return x_, y_
cp_1 = Trainer.checkpoint_tuple(
id='cp_1',
iterations=111,
model_state=SequentialLinear(50, 2).state_dict(),
)
cp_2 = Trainer.checkpoint_tuple(
id='cp_2',
iterations=111,
model_state=SequentialLinear(50, 2).state_dict(),
)
# save checkpoint
p = Persist('test_model_1', increment=False, only_best_states=False)
p.before_proc(trainer=_Trainer())
p.on_checkpoint(cp_1)
p.on_checkpoint(cp_2)
assert (Path('.').resolve() / 'test_model_1' / 'checkpoints' /
'cp_1.pth.s').exists()
assert (Path('.').resolve() / 'test_model_1' / 'checkpoints' /
'cp_2.pth.s').exists()
# reduced save checkpoint
p = Persist('test_model_2', increment=False, only_best_states=True)
p.before_proc(trainer=_Trainer())
p.on_checkpoint(cp_1)
p.on_checkpoint(cp_2)
assert (Path('.').resolve() / 'test_model_2' / 'checkpoints' /
'cp.pth.s').exists()
assert not (Path('.').resolve() / 'test_model_2' / 'checkpoints' /
'cp_1.pth.s').exists()
assert not (Path('.').resolve() / 'test_model_2' / 'checkpoints' /
'cp_2.pth.s').exists()
# no checkpoint will be saved
p = Persist('test_model_3', increment=False, only_best_states=True)
p.before_proc(trainer=_Trainer())
p.on_checkpoint(cp_2)
assert not (Path('.').resolve() / 'test_model_3' / 'checkpoints' /
'cp.pth.s').exists()
assert not (Path('.').resolve() / 'test_model_3' / 'checkpoints' /
'cp_2.pth.s').exists()
def test_sequential_1(data):
m = SequentialLinear(10, 5)
assert isinstance(m, torch.nn.Module)
assert m.output.in_features == 10
assert m.output.out_features == 5
x = m(data[0])
assert x.size() == (10, 5)
def test_sequential_3():
m = SequentialLinear(10, 1, bias=False, h_neurons=(0.7, 0.5))
assert m.layer_0.linear.in_features == 10
assert m.layer_0.linear.out_features == 7
assert m.layer_0.linear.bias is None
assert m.layer_1.linear.in_features == 7
assert m.layer_1.linear.out_features == 5
assert m.layer_1.linear.bias is not None
assert m.output.in_features == 5
assert m.output.out_features == 1
assert m.output.bias is not None
def test_sequential_2(data):
m = SequentialLinear(10, 1, bias=False, h_neurons=(7, 4))
assert isinstance(m.layer_0, LinearLayer)
assert isinstance(m.layer_1, LinearLayer)
assert m.layer_0.linear.in_features == 10
assert m.layer_0.linear.out_features == 7
assert m.layer_0.linear.bias is None
assert m.layer_1.linear.in_features == 7
assert m.layer_1.linear.out_features == 4
assert m.layer_1.linear.bias is not None
assert m.output.in_features == 4
assert m.output.out_features == 1
assert m.output.bias is not None
x = m(data[0])
assert x.size() == (10, 1)
def data():
# prepare path
model1 = nn.Sequential(
nn.Sequential(OrderedDict(layer=nn.Linear(10, 7), act_func=nn.ReLU())),
nn.Sequential(OrderedDict(layer=nn.Linear(7, 4), act_func=nn.ReLU())),
nn.Sequential(OrderedDict(layer=nn.Linear(4, 1))),
)
model2 = SequentialLinear(10, 1, h_neurons=(7, 4))
descriptor1 = np.random.rand(10, 10)
descriptor2 = np.random.rand(20, 10)
index = ['id_' + str(i) for i in range(10)]
df = pd.DataFrame(descriptor1, index=index)
# ignore numpy warning
import warnings
print('ignore NumPy RuntimeWarning\n')
warnings.filterwarnings("ignore", message="numpy.dtype size changed")
warnings.filterwarnings("ignore", message="numpy.ndarray size changed")
yield model1, descriptor1, df, index, descriptor2, model2
print('test over')
def __init__(self):
super().__init__()
self.model = SequentialLinear(50, 2)