def test_auto_should_be_max(self):
state = {
torchbearer.EPOCH: 1,
torchbearer.STOP_TRAINING: False,
torchbearer.METRICS: {'acc_metric': 0.001}
}
stopper = EarlyStopping(monitor='acc_metric')
stopper.on_start(state)
stopper.on_end_epoch(state)
self.assertTrue(stopper.mode == 'max')
def test_state_dict(self):
stopper = EarlyStopping(monitor='test_metric_1')
stopper.wait = 10
stopper.best = 20
state = stopper.state_dict()
stopper = EarlyStopping(monitor='test_metric_1')
self.assertNotEqual(stopper.wait, 10)
stopper.load_state_dict(state)
self.assertEqual(stopper.wait, 10)
self.assertEqual(stopper.best, 20)
def test_step_on_batch(self):
stopper = EarlyStopping(monitor='test_metric',
mode='min',
step_on_batch=True)
stopper.step = MagicMock()
stopper.on_step_training('test')
self.assertTrue(stopper.step.call_count == 1)
stopper.on_end_epoch('test')
self.assertTrue(stopper.step.call_count == 1)
def test_max_equal_should_stop(self):
state = {
torchbearer.EPOCH: 1,
torchbearer.STOP_TRAINING: False,
torchbearer.METRICS: {
'test_metric': 0.001
}
}
stopper = EarlyStopping(monitor='test_metric', mode='max')
stopper.on_start(state)
stopper.on_end_epoch(state)
self.assertFalse(state[torchbearer.STOP_TRAINING])
stopper.on_end_epoch(state)
self.assertTrue(state[torchbearer.STOP_TRAINING])
def test_patience_should_stop(self):
state = {
torchbearer.EPOCH: 1,
torchbearer.STOP_TRAINING: False,
torchbearer.METRICS: {
'test_metric': 0.001
}
}
stopper = EarlyStopping(monitor='test_metric', patience=3)
stopper.on_start(state)
for i in range(3):
stopper.on_end_epoch(state)
self.assertFalse(state[torchbearer.STOP_TRAINING])
stopper.on_end_epoch(state)
self.assertTrue(state[torchbearer.STOP_TRAINING])
def test_max_equal_should_stop(self):
state = {
torchbearer.EPOCH: 1,
torchbearer.STOP_TRAINING: False,
torchbearer.METRICS: {'test_metric': 0.001}
}
stopper = EarlyStopping(monitor='test_metric', mode='max')
stopper.on_start(state)
stopper.on_end_epoch(state)
self.assertFalse(state[torchbearer.STOP_TRAINING])
stopper.on_end_epoch(state)
self.assertTrue(state[torchbearer.STOP_TRAINING])
def test_patience_should_stop(self):
state = {
torchbearer.EPOCH: 1,
torchbearer.STOP_TRAINING: False,
torchbearer.METRICS: {'test_metric': 0.001}
}
stopper = EarlyStopping(monitor='test_metric', patience=3)
stopper.on_start(state)
for i in range(3):
stopper.on_end_epoch(state)
self.assertFalse(state[torchbearer.STOP_TRAINING])
stopper.on_end_epoch(state)
self.assertTrue(state[torchbearer.STOP_TRAINING])
def test_min_should_continue(self):
state = {
torchbearer.EPOCH: 1,
torchbearer.STOP_TRAINING: False,
torchbearer.METRICS: {
'test_metric': 0.001
}
}
stopper = EarlyStopping(monitor='test_metric', mode='min')
stopper.on_start(state)
stopper.on_end_epoch(state)
self.assertFalse(state[torchbearer.STOP_TRAINING])
state[torchbearer.METRICS]['test_metric'] = 0.0001
stopper.on_end_epoch(state)
self.assertFalse(state[torchbearer.STOP_TRAINING])
def test_min_delta_should_continue(self):
state = {
torchbearer.EPOCH: 1,
torchbearer.STOP_TRAINING: False,
torchbearer.METRICS: {'test_metric': 0.001}
}
stopper = EarlyStopping(monitor='test_metric', mode='max', min_delta=0.1)
stopper.on_start(state)
stopper.on_end_epoch(state)
self.assertFalse(state[torchbearer.STOP_TRAINING])
state[torchbearer.METRICS]['test_metric'] = 0.102
stopper.on_end_epoch(state)
self.assertFalse(state[torchbearer.STOP_TRAINING])
def test_auto_should_be_max(self):
state = {
torchbearer.EPOCH: 1,
torchbearer.STOP_TRAINING: False,
torchbearer.METRICS: {
'acc_metric': 0.001
}
}
stopper = EarlyStopping(monitor='acc_metric')
stopper.on_start(state)
stopper.on_end_epoch(state)
self.assertTrue(stopper.mode == 'max')
