def test_validator_1(data):
x = data[0]
y = data[1]
class _Trainer(BaseRunner):
def __init__(self):
super().__init__()
self.x_val = x
self.y_val = y
self.loss_type = 'train_loss'
def predict(self, x_, y_):
return x_, y_
val = Validator(metrics_func=regression_metrics, each_iteration=False)
step_info = OrderedDict(train_loss=0, i_epoch=0)
val.step_forward(trainer=_Trainer(), step_info=step_info)
assert 'val_mae' not in step_info
step_info = OrderedDict(train_loss=0, i_epoch=1)
val.step_forward(trainer=_Trainer(), step_info=step_info)
assert step_info['val_mae'] == regression_metrics(x, y)['mae']
assert set(step_info.keys()) == {
'i_epoch', 'val_mae', 'val_mse', 'val_rmse', 'val_r2', 'val_pearsonr',
'val_spearmanr', 'val_p_value', 'val_max_ae', 'train_loss'
}
def test_validator_1():
x = np.random.randn(100) # input
y = x + np.random.rand() * 0.001 # true values
class _Trainer(BaseRunner):
def __init__(self):
super().__init__()
self.x_val = x
self.y_val = y
self.loss_type = 'train_loss'
def predict(self, x_, y_):
return x_, y_
val = Validator('regress', each_iteration=False)
step_info = OrderedDict(train_loss=0, i_epoch=0)
val.step_forward(trainer=_Trainer(), step_info=step_info) # noqa
assert 'val_mae' not in step_info
step_info = OrderedDict(train_loss=0, i_epoch=1)
val.step_forward(trainer=_Trainer(), step_info=step_info) # noqa
assert step_info['val_mae'] == regression_metrics(y, x)['mae']
assert set(step_info.keys()) == {
'i_epoch', 'val_mae', 'val_mse', 'val_rmse', 'val_r2', 'val_pearsonr', 'val_spearmanr',
'val_p_value', 'val_max_ae', 'train_loss'
}
def test_regression_metrics_1(data):
x = data[0]
y = data[1]
noise = data[2]
metric = regression_metrics(x, y)
assert isinstance(metric, dict)
assert set(metric.keys()) == {'mae', 'mse', 'rmse', 'r2', 'pearsonr', 'spearmanr', 'p_value', 'max_ae'}
assert metric['mae'] < noise
assert metric['mse'] < noise ** 2
assert metric['rmse'] < noise
assert metric['r2'] > 0.9999
assert metric['pearsonr'] > 0.9999
assert metric['spearmanr'] > 0.9999
assert np.isclose(metric['p_value'], 0, 1e-4)
assert metric['max_ae'] < noise
assert metric['mae'] == regression_metrics(x.reshape(-1, 1), y)['mae']
assert metric['rmse'] == regression_metrics(x.reshape(-1, 1), y.reshape(-1, 1))['rmse']
assert metric['max_ae'] == regression_metrics(x, y.reshape(-1, 1))['max_ae']