def _test_epoch_metric(self, pred, true, sklearn_metrics, with_sample_weight, *, kwargs=None, names=None):
epoch_metric = SKLearnMetrics(sklearn_metrics, kwargs=kwargs, names=names)
if with_sample_weight:
loader = self._get_data_loader(pred, (true, self.sample_weight))
sample_weight = self.sample_weight.numpy()
else:
loader = self._get_data_loader(pred, true)
sample_weight = None
true = true.numpy()
pred = pred.numpy()
if not isinstance(sklearn_metrics, list):
sklearn_metrics = [sklearn_metrics]
if kwargs is not None and not isinstance(kwargs, list):
kwargs = [kwargs]
kwargs = kwargs if kwargs is not None else repeat({})
if names is not None and not isinstance(names, list):
names = [names]
names = [func.__name__ for func in sklearn_metrics] if names is None else names
expected = {
name: f(true, pred, sample_weight=sample_weight, **kw)
for name, f, kw in zip(names, sklearn_metrics, kwargs)
}
with torch.no_grad():
for y_pred, y_true in loader:
epoch_metric(y_pred, y_true)
actual = epoch_metric.get_metric()
self.assertEqual(expected, actual)
test_x = np.random.randn(num_test_samples, num_features).astype('float32')
test_y = np.random.randn(num_test_samples, 1).astype('float32')
cuda_device = 0
device = torch.device("cuda:%d" %
cuda_device if torch.cuda.is_available() else "cpu")
# Define the network
network = nn.Sequential(
nn.Linear(num_features, hidden_state_size),
nn.ReLU(),
nn.Linear(hidden_state_size, 1),
)
# Train
model = Model(
network,
'sgd',
'mse',
batch_metrics=['l1'],
epoch_metrics=[SKLearnMetrics(r2_score)],
device=device,
)
model.fit(
train_x,
train_y,
validation_data=(valid_x, valid_y),
epochs=5,
batch_size=32,
)