def __call__(self, nn, train_history):
current_valid = train_history[-1][self.loss] \
* (-1.0 if self.greater_is_better else 1.0)
current_epoch = train_history[-1]['epoch']
if current_valid < self.best_valid:
self.best_valid = current_valid
self.best_valid_epoch = current_epoch
self.best_weights = [w.get_value() for w in nn.get_all_params()]
nn.save_params_to(self.weights_file)
def __call__(self, nn, train_history):
current_valid = train_history[-1][self.loss] \
* (-1.0 if self.greater_is_better else 1.0)
current_epoch = train_history[-1]['epoch']
if current_valid < self.best_valid:
self.best_valid = current_valid
self.best_valid_epoch = current_epoch
self.best_weights = [w.get_value() for w in nn.get_all_params()]
nn.save_params_to(self.weights_file)
def __call__(self, nn, train_history):
epoch = train_history[-1]['epoch']
if epoch in self.schedule:
new_value = self.schedule[epoch]
if new_value == 'stop':
if self.weights_file is not None:
nn.save_params_to(self.weights_file)
raise StopIteration
getattr(nn, self.name).set_value(util.float32(new_value))
def __call__(self, nn, train_history):
epoch = train_history[-1]['epoch']
if epoch in self.schedule:
new_value = self.schedule[epoch]
if new_value == 'stop':
if self.weights_file is not None:
nn.save_params_to(self.weights_file)
raise StopIteration
getattr(nn, self.name).set_value(util.float32(new_value))
