def test_r2_sklearn_comparison(multioutput):
"""Test against sklearn's implementation on random inputs."""
for _ in range(10):
actuals = np.random.rand(64, 3)
preds = np.random.rand(64, 3)
sample_weight = np.random.rand(64, 1)
tensor_actuals = tf.constant(actuals, dtype=tf.float32)
tensor_preds = tf.constant(preds, dtype=tf.float32)
tensor_sample_weight = tf.constant(sample_weight, dtype=tf.float32)
tensor_actuals = tf.cast(tensor_actuals, dtype=tf.float32)
tensor_preds = tf.cast(tensor_preds, dtype=tf.float32)
tensor_sample_weight = tf.cast(tensor_sample_weight, dtype=tf.float32)
# Initialize
r2_obj = initialize_vars(y_shape=(3, ), multioutput=multioutput)
# Update
update_obj_states(
r2_obj,
tensor_actuals,
tensor_preds,
sample_weight=tensor_sample_weight,
)
# Check results by comparing to results of scikit-learn r2 implementation
sklearn_result = sklearn_r2_score(actuals,
preds,
sample_weight=sample_weight,
multioutput=multioutput)
check_results(r2_obj, sklearn_result)
def test_r2_sklearn_comparison():
"""Test that RSquare behaves similarly to the scikit-learn
implementation of the same metric, given random input.
"""
for multioutput in VALID_MULTIOUTPUT:
for _ in range(10):
actuals = np.random.rand(64, 3)
preds = np.random.rand(64, 3)
sample_weight = np.random.rand(64, 1)
tensor_actuals = tf.constant(actuals, dtype=tf.float32)
tensor_preds = tf.constant(preds, dtype=tf.float32)
tensor_sample_weight = tf.constant(sample_weight, dtype=tf.float32)
tensor_actuals = tf.cast(tensor_actuals, dtype=tf.float32)
tensor_preds = tf.cast(tensor_preds, dtype=tf.float32)
tensor_sample_weight = tf.cast(tensor_sample_weight,
dtype=tf.float32)
# Initialize
r2_obj = initialize_vars(y_shape=(3, ), multioutput=multioutput)
# Update
update_obj_states(
r2_obj,
tensor_actuals,
tensor_preds,
sample_weight=tensor_sample_weight,
)
# Check results by comparing to results of scikit-learn r2 implementation
sklearn_result = sklearn_r2_score(actuals,
preds,
sample_weight=sample_weight,
multioutput=multioutput)
check_results(r2_obj, sklearn_result)
def test_r2_sklearn_comparison(self):
"""Test that RSquare behaves similarly to the scikit-learn
implementation of the same metric, given random input.
"""
for i in range(10):
actuals = np.random.rand(64, 1)
preds = np.random.rand(64, 1)
sample_weight = np.random.rand(64, 1)
tensor_actuals = tf.constant(actuals, dtype=tf.float32)
tensor_preds = tf.constant(preds, dtype=tf.float32)
tensor_sample_weight = tf.constant(sample_weight, dtype=tf.float32)
tensor_actuals = tf.cast(tensor_actuals, dtype=tf.float32)
tensor_preds = tf.cast(tensor_preds, dtype=tf.float32)
tensor_sample_weight = tf.cast(tensor_sample_weight,
dtype=tf.float32)
# Initialize
r2_obj = self.initialize_vars()
# Update
self.update_obj_states(r2_obj,
tensor_actuals,
tensor_preds,
sample_weight=tensor_sample_weight)
# Check results by comparing to results of scikit-learn r2 implementation
sklearn_result = sklearn_r2_score(actuals,
preds,
sample_weight=sample_weight)
self.check_results(r2_obj, sklearn_result)
def r2_score(model_output: Union[torch.Tensor, np.ndarray], label: Union[torch.Tensor, np.ndarray]) \
-> float:
"""
Computes the coefficient of determination R2. Represents the proportion of variance explained
by the (independent) variables in the model. R2 = 1 - Mean(SquaredErrors) / Variance(Labels)
"""
if isinstance(label, torch.Tensor):
label = label.detach().cpu().numpy()
if isinstance(model_output, torch.Tensor):
model_output = model_output.detach().cpu().numpy()
return sklearn_r2_score(label, model_output)
def scorer(y_true, y_pred, **kwargs) -> float:
"""R^2 score based on true and predicted target values.
Parameters
----------
y_true : array-like
True labels.
y_pred : array-like
Predicted labels.
Returns
-------
score
float
"""
return sklearn_r2_score(y_true, y_pred, **kwargs)
def test_kerasregressor_r2():
"""Test custom R^2 implementation against scikit-learn's."""
n_samples = 50
datasets = []
y_true = np.arange(n_samples, dtype=float)
y_pred = y_true + 1
datasets.append((y_true.reshape(-1, 1), y_pred.reshape(-1, 1)))
y_true = np.random.random_sample(size=y_true.shape)
y_pred = np.random.random_sample(size=y_true.shape)
datasets.append((y_true.reshape(-1, 1), y_pred.reshape(-1, 1)))
def keras_backend_r2(y_true, y_pred):
"""Wrap Keras operations to numpy."""
y_true = convert_to_tensor(y_true)
y_pred = convert_to_tensor(y_pred)
return KerasRegressor.r_squared(y_true, y_pred).numpy()
for (y_true, y_pred) in datasets:
np.testing.assert_almost_equal(
keras_backend_r2(y_true, y_pred),
sklearn_r2_score(y_true, y_pred),
decimal=5,
)
def r2_score(y_true, y_pred):
from sklearn.metrics import r2_score as sklearn_r2_score
y_true = convert_to_numpy(y_true)
y_pred = convert_to_numpy(y_pred)
return sklearn_r2_score(y_true, y_pred)
