class NonTerminalTerminalAccuracyMetrics(Metrics):
def __init__(self):
super().__init__()
self.nt_accuracy = MaxPredictionAccuracyMetrics()
self.t_accuracy = MaxPredictionAccuracyMetrics()
def drop_state(self):
self.nt_accuracy.drop_state()
self.t_accuracy.drop_state()
def report(self, data):
nt_prediction, t_prediction, nt_target, t_target = data
self.nt_accuracy.report((nt_prediction, nt_target))
self.t_accuracy.report((t_prediction, t_target))
def get_current_value(self, should_print=False):
nt_value = self.nt_accuracy.get_current_value(should_print=False)
t_value = self.t_accuracy.get_current_value(should_print=False)
if should_print:
print('Non terminals accuracy: {}'.format(nt_value))
print('Terminals accuracy: {}'.format(t_value))
return nt_value, t_value
def create_eval_metrics(self, args):
return SequentialMetrics([
NonTerminalMetrics(base=MaxPredictionAccuracyMetrics()),
SingleNonTerminalAccuracyMetrics(
non_terminals_file=args.non_terminals_file,
results_dir=args.eval_results_directory
),
NonTerminalsMetricsWrapper(TopKWrapper(base=ResultsSaver(dir_to_save=args.eval_results_directory)))
])
def print_results(args):
# assert args.prediction == 'nt2n'
# seed = 1000
# random.seed(seed)
# numpy.random.seed(seed)
main = get_main(args)
routine = main.validation_routine
metrics = MaxPredictionAccuracyMetrics()
metrics.drop_state()
main.model.eval()
for iter_num, iter_data in enumerate(
tqdm_lim(main.data_generator.get_eval_generator(), lim=1000)):
metrics_data = routine.run(iter_num, iter_data)
metrics.report(metrics_data)
metrics.get_current_value(should_print=True)
def create_eval_metrics(self, args) -> Metrics:
return SequentialMetrics([
MaxPredictionAccuracyMetrics(),
TopKWrapper(base=ResultsSaver(dir_to_save=args.eval_results_directory))
])
def create_train_metrics(self, args) -> Metrics:
return MaxPredictionAccuracyMetrics()
def create_train_metrics(self, args):
return TerminalMetrics(base=MaxPredictionAccuracyMetrics())
def __init__(self):
super().__init__()
self.nt_accuracy = MaxPredictionAccuracyMetrics()
self.t_accuracy = MaxPredictionAccuracyMetrics()