51 lines
1.5 KiB
Python
51 lines
1.5 KiB
Python
|
from pytorch_lightning import Trainer
|
||
|
|
from tests.base.boring_model import BoringModel
|
||
|
|
import torch
|
||
|
|
|
||
|
|
|
||
|
|
def test_multiple_optimizers(tmpdir):
|
||
|
|
"""
|
||
|
|
Tests that only training_step can be used
|
||
|
|
"""
|
||
|
|
class TestModel(BoringModel):
|
||
|
|
def on_train_epoch_start(self) -> None:
|
||
|
|
self.opt_0_seen = False
|
||
|
|
self.opt_1_seen = False
|
||
|
|
|
||
|
|
def training_step(self, batch, batch_idx, optimizer_idx):
|
||
|
|
if optimizer_idx == 0:
|
||
|
|
self.opt_0_seen = True
|
||
|
|
elif optimizer_idx == 1:
|
||
|
|
self.opt_1_seen = True
|
||
|
|
else:
|
||
|
|
raise Exception('should only have two optimizers')
|
||
|
|
|
||
|
|
self.training_step_called = True
|
||
|
|
loss = self.step(batch[0])
|
||
|
|
return loss
|
||
|
|
|
||
|
|
def training_epoch_end(self, outputs) -> None:
|
||
|
|
# outputs should be an array with an entry per optimizer
|
||
|
|
assert len(outputs) == 2
|
||
|
|
|
||
|
|
def configure_optimizers(self):
|
||
|
|
optimizer = torch.optim.SGD(self.layer.parameters(), lr=0.1)
|
||
|
|
optimizer_2 = torch.optim.SGD(self.layer.parameters(), lr=0.1)
|
||
|
|
return optimizer, optimizer_2
|
||
|
|
|
||
|
|
model = TestModel()
|
||
|
|
model.val_dataloader = None
|
||
|
|
|
||
|
|
trainer = Trainer(
|
||
|
|
default_root_dir=tmpdir,
|
||
|
|
limit_train_batches=2,
|
||
|
|
limit_val_batches=2,
|
||
|
|
max_epochs=1,
|
||
|
|
row_log_interval=1,
|
||
|
|
weights_summary=None,
|
||
|
|
)
|
||
|
|
|
||
|
|
trainer.fit(model)
|
||
|
|
assert model.opt_0_seen
|
||
|
|
assert model.opt_1_seen
|