def _on_epoch_end(self):
self.on_epoch_end()
epoch, phase = self.epoch_and_phase
self.fire(Event.ON_EPOCH_END(Moment(epoch, Frequency.EPOCH, phase)),
self.epoch_monitors(phase))
def _on_epoch_begin(self):
self._reset_model_trainers()
self.on_epoch_begin()
epoch, phase = self.epoch_and_phase
self.fire(Event.ON_EPOCH_BEGIN(Moment(epoch, Frequency.EPOCH, phase)))
def _on_batch_begin(self):
self.on_batch_begin()
iteration, phase = self.iteration_and_phase
self.fire(
Event.ON_BATCH_BEGIN(Moment(iteration, Frequency.STEP, phase)))
def _on_batch_end(self):
self.on_batch_end()
iteration, phase = self.iteration_and_phase
self.fire(Event.ON_BATCH_END(Moment(iteration, Frequency.STEP, phase)),
self.step_monitors(phase))
def _on_valid_batch_end(self):
self.on_valid_batch_begin()
self.fire(
Event.ON_VALID_BATCH_END(
Moment(self.current_valid_step,
Frequency.STEP, Phase.VALIDATION)),
self.step_monitors(Phase.VALIDATION))
def test_should_not_save_model_with_higher_valid_losses(
self, model_state_mock, optimizer_states_mock):
model_state_mock.return_value = dict()
optimizer_states_mock.return_value = list(dict())
moment = Moment(200, Frequency.EPOCH, Phase.VALIDATION)
handler_mock = mockito.spy(
Checkpoint(self.SAVE_PATH, self.MODEL_NAME, ["MSELoss", "L1Loss"],
0.01, MonitorMode.MIN))
monitors = {
self.MODEL_NAME: {
Phase.TRAINING: {
Monitor.METRICS: {},
Monitor.LOSS: {}
},
Phase.VALIDATION: {
Monitor.METRICS: {},
Monitor.LOSS: {
"MSELoss": torch.tensor([0.5]),
"L1Loss": torch.tensor([0.5])
}
},
Phase.TEST: {
Monitor.METRICS: {},
Monitor.LOSS: {}
}
}
}
handler_mock(TemporalEvent(Event.ON_EPOCH_END, moment), monitors,
self._trainer_mock)
monitors = {
self.MODEL_NAME: {
Phase.TRAINING: {
Monitor.METRICS: {},
Monitor.LOSS: {}
},
Phase.VALIDATION: {
Monitor.METRICS: {},
Monitor.LOSS: {
"MSELoss": torch.tensor([0.5]),
"L1Loss": torch.tensor([0.6])
}
},
Phase.TEST: {
Monitor.METRICS: {},
Monitor.LOSS: {}
}
}
}
handler_mock(TemporalEvent(Event.ON_EPOCH_END, moment), monitors,
self._trainer_mock)
assert_that(not os.path.exists(
os.path.join(self.SAVE_PATH, self.MODEL_NAME, self.MODEL_NAME +
".tar")))
def _on_test_epoch_end(self):
self._current_valid_batch = 0
self._current_test_batch = 0
self.on_test_epoch_end()
self.fire(
Event.ON_TEST_EPOCH_END(
Moment(self.epoch, Frequency.STEP, Phase.TEST)),
self.epoch_monitors(Phase.TEST))
def _on_valid_batch_begin(self):
self.on_valid_batch_begin()
self.fire(
Event.ON_VALID_BATCH_BEGIN(
Moment(self.current_valid_step, Frequency.STEP,
Phase.VALIDATION)))
def _on_train_batch_begin(self):
self.on_train_batch_begin()
self.fire(
Event.ON_TRAIN_BATCH_BEGIN(
Moment(self.current_train_step, Frequency.STEP,
Phase.TRAINING)))
def _on_train_batch_end(self):
self.on_train_batch_end()
self.fire(
Event.ON_TRAIN_BATCH_END(
Moment(self.current_train_step, Frequency.STEP,
Phase.TRAINING)), self.step_monitors(Phase.TRAINING))
def _on_test_batch_end(self):
self.on_test_batch_end()
self.fire(
Event.ON_TEST_BATCH_END(
Moment(self.current_test_step, Frequency.STEP, Phase.TEST)),
self.step_monitors(Phase.TEST))
def _on_test_batch_begin(self):
self.on_test_batch_begin()
self.fire(
Event.ON_TEST_BATCH_BEGIN(
Moment(self.current_test_step, Frequency.STEP, Phase.TEST)))
def _on_test_begin(self):
self.on_test_begin()
self._status = Status.TESTING
self.fire(Event.ON_TEST_BEGIN(Moment(0, Frequency.PHASE, Phase.TEST)))
def _on_test_end(self):
self.on_test_end()
self.fire(Event.ON_TEST_END(Moment(0, Frequency.PHASE, Phase.TEST)))
def _on_valid_begin(self):
self.on_valid_begin()
self._status = Status.VALIDATING
self.fire(
Event.ON_VALID_BEGIN(Moment(0, Frequency.PHASE, Phase.VALIDATION)))
def _on_valid_end(self):
self.on_valid_end()
self.fire(
Event.ON_VALID_END(Moment(0, Frequency.PHASE, Phase.VALIDATION)))
def _on_training_begin(self):
self.on_training_begin()
self._status = Status.TRAINING
self.fire(
Event.ON_TRAINING_BEGIN(Moment(0, Frequency.PHASE,
Phase.TRAINING)))
def _on_training_end(self):
self.on_training_end()
self.scheduler_step()
self.fire(
Event.ON_TRAINING_END(Moment(0, Frequency.PHASE, Phase.TRAINING)))
def _on_train_epoch_begin(self):
self._status = Status.TRAINING
self.on_train_epoch_begin()
self.fire(
Event.ON_TRAIN_EPOCH_BEGIN(
Moment(self.epoch, Frequency.EPOCH, Phase.TRAINING)))
def _on_test_epoch_begin(self):
self.on_test_epoch_begin()
self.fire(
Event.ON_TEST_EPOCH_BEGIN(
Moment(self.epoch, Frequency.EPOCH, Phase.TEST)))
def _on_valid_epoch_end(self):
self.on_valid_epoch_end()
self.fire(
Event.ON_VALID_EPOCH_END(
Moment(self.epoch, Frequency.EPOCH, Phase.VALIDATION)),
self.epoch_monitors(Phase.VALIDATION))
def _on_valid_epoch_begin(self):
self.on_valid_epoch_begin()
self.fire(
Event.ON_VALID_EPOCH_BEGIN(
Moment(self.epoch, Frequency.EPOCH, Phase.VALIDATION)))
def _on_train_epoch_end(self):
self.on_train_epoch_end()
self.fire(
Event.ON_TRAIN_EPOCH_END(
Moment(self.epoch, Frequency.EPOCH, Phase.TRAINING)),
self.epoch_monitors(Phase.TRAINING))