def test_multinomial_loss():
# test if the multinomial loss and gradient computations are consistent
X, y = iris.data, iris.target.astype(np.float64)
n_samples, n_features = X.shape
n_classes = len(np.unique(y))
rng = check_random_state(42)
weights = rng.randn(n_features, n_classes)
intercept = rng.randn(n_classes)
sample_weights = rng.randn(n_samples)
np.abs(sample_weights, sample_weights)
# compute loss and gradient like in multinomial SAG
dataset, _ = make_dataset(X, y, sample_weights, random_state=42)
loss_1, grad_1 = _multinomial_grad_loss_all_samples(
dataset, weights, intercept, n_samples, n_features, n_classes)
# compute loss and gradient like in multinomial LogisticRegression
lbin = LabelBinarizer()
Y_bin = lbin.fit_transform(y)
weights_intercept = np.vstack((weights, intercept)).T.ravel()
loss_2, grad_2, _ = _multinomial_loss_grad(weights_intercept, X, Y_bin,
0.0, sample_weights)
grad_2 = grad_2.reshape(n_classes, -1)
grad_2 = grad_2[:, :-1].T
# comparison
assert_array_almost_equal(grad_1, grad_2)
assert_almost_equal(loss_1, loss_2)
def test_multinomial_loss():
# test if the multinomial loss and gradient computations are consistent
X, y = iris.data, iris.target.astype(np.float64)
n_samples, n_features = X.shape
n_classes = len(np.unique(y))
rng = check_random_state(42)
weights = rng.randn(n_features, n_classes)
intercept = rng.randn(n_classes)
sample_weights = rng.randn(n_samples)
np.abs(sample_weights, sample_weights)
# compute loss and gradient like in multinomial SAG
dataset, _ = make_dataset(X, y, sample_weights, random_state=42)
loss_1, grad_1 = _multinomial_grad_loss_all_samples(dataset, weights,
intercept, n_samples,
n_features, n_classes)
# compute loss and gradient like in multinomial LogisticRegression
lbin = LabelBinarizer()
Y_bin = lbin.fit_transform(y)
weights_intercept = np.vstack((weights, intercept)).T.ravel()
loss_2, grad_2, _ = _multinomial_loss_grad(weights_intercept, X, Y_bin,
0.0, sample_weights)
grad_2 = grad_2.reshape(n_classes, -1)
grad_2 = grad_2[:, :-1].T
# comparison
assert_array_almost_equal(grad_1, grad_2)
assert_almost_equal(loss_1, loss_2)
def test_fused_types_make_dataset():
iris = load_iris()
X_32 = iris.data.astype(np.float32)
y_32 = iris.target.astype(np.float32)
X_csr_32 = sparse.csr_matrix(X_32)
sample_weight_32 = np.arange(y_32.size, dtype=np.float32)
X_64 = iris.data.astype(np.float64)
y_64 = iris.target.astype(np.float64)
X_csr_64 = sparse.csr_matrix(X_64)
sample_weight_64 = np.arange(y_64.size, dtype=np.float64)
# array
dataset_32, _ = make_dataset(X_32, y_32, sample_weight_32)
dataset_64, _ = make_dataset(X_64, y_64, sample_weight_64)
xi_32, yi_32, _, _ = dataset_32._next_py()
xi_64, yi_64, _, _ = dataset_64._next_py()
xi_data_32, _, _ = xi_32
xi_data_64, _, _ = xi_64
assert_equal(xi_data_32.dtype, np.float32)
assert_equal(xi_data_64.dtype, np.float64)
assert isinstance(yi_32, float)
assert isinstance(yi_64, float)
# assert_array_almost_equal(yi_64, yi_32, decimal=5)
# csr
datasetcsr_32, _ = make_dataset(X_csr_32, y_32, sample_weight_32)
datasetcsr_64, _ = make_dataset(X_csr_64, y_64, sample_weight_64)
xicsr_32, yicsr_32, _, _ = datasetcsr_32._next_py()
xicsr_64, yicsr_64, _, _ = datasetcsr_64._next_py()
xicsr_data_32, _, _ = xicsr_32
xicsr_data_64, _, _ = xicsr_64
assert_equal(xicsr_data_32.dtype, np.float32)
assert_equal(xicsr_data_64.dtype, np.float64)
assert isinstance(yicsr_32, float)
assert isinstance(yicsr_64, float)
assert_array_almost_equal(xicsr_data_64, xicsr_data_32, decimal=5)
assert_array_almost_equal(yicsr_64, yicsr_32, decimal=5)
assert_array_equal(xi_data_32, xicsr_data_32)
assert_array_equal(xi_data_64, xicsr_data_64)
assert_array_equal(yi_32, yicsr_32)
assert_array_equal(yi_64, yicsr_64)
def test_fused_types_make_dataset():
iris = load_iris()
X_32 = iris.data.astype(np.float32)
y_32 = iris.target.astype(np.float32)
X_csr_32 = sparse.csr_matrix(X_32)
sample_weight_32 = np.arange(y_32.size, dtype=np.float32)
X_64 = iris.data.astype(np.float64)
y_64 = iris.target.astype(np.float64)
X_csr_64 = sparse.csr_matrix(X_64)
sample_weight_64 = np.arange(y_64.size, dtype=np.float64)
# array
dataset_32, _ = make_dataset(X_32, y_32, sample_weight_32)
dataset_64, _ = make_dataset(X_64, y_64, sample_weight_64)
xi_32, yi_32, _, _ = dataset_32._next_py()
xi_64, yi_64, _, _ = dataset_64._next_py()
xi_data_32, _, _ = xi_32
xi_data_64, _, _ = xi_64
assert xi_data_32.dtype == np.float32
assert xi_data_64.dtype == np.float64
assert_allclose(yi_64, yi_32, rtol=rtol)
# csr
datasetcsr_32, _ = make_dataset(X_csr_32, y_32, sample_weight_32)
datasetcsr_64, _ = make_dataset(X_csr_64, y_64, sample_weight_64)
xicsr_32, yicsr_32, _, _ = datasetcsr_32._next_py()
xicsr_64, yicsr_64, _, _ = datasetcsr_64._next_py()
xicsr_data_32, _, _ = xicsr_32
xicsr_data_64, _, _ = xicsr_64
assert xicsr_data_32.dtype == np.float32
assert xicsr_data_64.dtype == np.float64
assert_allclose(xicsr_data_64, xicsr_data_32, rtol=rtol)
assert_allclose(yicsr_64, yicsr_32, rtol=rtol)
assert_array_equal(xi_data_32, xicsr_data_32)
assert_array_equal(xi_data_64, xicsr_data_64)
assert_array_equal(yi_32, yicsr_32)
assert_array_equal(yi_64, yicsr_64)
def _fit_regressor(self, X, y, alpha, C, loss, learning_rate,
sample_weight, n_iter):
dataset, intercept_decay = make_dataset(X, y, sample_weight)
self.coef_ = np.zeros((3,), dtype=np.float64, order="C")
loss_function = self._get_loss_function(loss)
penalty_type = self._get_penalty_type(self.penalty)
learning_rate_type = self._get_learning_rate_type(learning_rate)
if self.t_ is None:
self.t_ = 1.0
random_state = check_random_state(self.random_state)
# numpy mtrand expects a C long which is a signed 32 bit integer under
# Windows
seed = random_state.randint(0, np.iinfo(np.int32).max)
if self.average > 0:
self.standard_coef_, self.standard_intercept_, \
self.average_coef_, self.average_intercept_ = \
average_sgd(self.standard_coef_,
self.standard_intercept_[0],
self.average_coef_,
self.average_intercept_[0],
loss_function,
penalty_type,
alpha, C,
self.l1_ratio,
dataset,
n_iter,
int(self.fit_intercept),
int(self.verbose),
int(self.shuffle),
seed,
1.0, 1.0,
learning_rate_type,
self.eta0, self.power_t, self.t_,
intercept_decay, self.average)
self.average_intercept_ = np.atleast_1d(self.average_intercept_)
self.standard_intercept_ = np.atleast_1d(self.standard_intercept_)
self.t_ += n_iter * X.shape[0]
if self.average <= self.t_ - 1.0:
self.coef_ = self.average_coef_
self.intercept_ = self.average_intercept_
else:
self.coef_ = self.standard_coef_
self.intercept_ = self.standard_intercept_
else:
self.coef_, self.intercept_ = \
self.parallelizer(self.coef_,
0.0,
loss_function,
penalty_type,
alpha, C,
self.l1_ratio,
dataset,
n_iter,
int(self.fit_intercept),
int(self.verbose),
int(self.shuffle),
seed,
1.0, 1.0,
learning_rate_type,
self.eta0, self.power_t, self.t_,
intercept_decay)
print(self.coef_)
print(self.intercept_)
self.t_ += n_iter * X.shape[0]
self.intercept_ = np.atleast_1d(self.intercept_)
def _fit_regressor(self, X, y, alpha, C, loss, learning_rate,
sample_weight, n_iter):
dataset, intercept_decay = make_dataset(X, y, sample_weight)
self.coef_ = np.zeros((3, ), dtype=np.float64, order="C")
loss_function = self._get_loss_function(loss)
penalty_type = self._get_penalty_type(self.penalty)
learning_rate_type = self._get_learning_rate_type(learning_rate)
if self.t_ is None:
self.t_ = 1.0
random_state = check_random_state(self.random_state)
# numpy mtrand expects a C long which is a signed 32 bit integer under
# Windows
seed = random_state.randint(0, np.iinfo(np.int32).max)
if self.average > 0:
self.standard_coef_, self.standard_intercept_, \
self.average_coef_, self.average_intercept_ = \
average_sgd(self.standard_coef_,
self.standard_intercept_[0],
self.average_coef_,
self.average_intercept_[0],
loss_function,
penalty_type,
alpha, C,
self.l1_ratio,
dataset,
n_iter,
int(self.fit_intercept),
int(self.verbose),
int(self.shuffle),
seed,
1.0, 1.0,
learning_rate_type,
self.eta0, self.power_t, self.t_,
intercept_decay, self.average)
self.average_intercept_ = np.atleast_1d(self.average_intercept_)
self.standard_intercept_ = np.atleast_1d(self.standard_intercept_)
self.t_ += n_iter * X.shape[0]
if self.average <= self.t_ - 1.0:
self.coef_ = self.average_coef_
self.intercept_ = self.average_intercept_
else:
self.coef_ = self.standard_coef_
self.intercept_ = self.standard_intercept_
else:
self.coef_, self.intercept_ = \
self.parallelizer(self.coef_,
0.0,
loss_function,
penalty_type,
alpha, C,
self.l1_ratio,
dataset,
n_iter,
int(self.fit_intercept),
int(self.verbose),
int(self.shuffle),
seed,
1.0, 1.0,
learning_rate_type,
self.eta0, self.power_t, self.t_,
intercept_decay)
print(self.coef_)
print(self.intercept_)
self.t_ += n_iter * X.shape[0]
self.intercept_ = np.atleast_1d(self.intercept_)