def test_sgd_default(dtype, datatype):
X, y = make_blobs(n_samples=100, n_features=3, centers=2, random_state=0)
X = X.astype(dtype)
y = y.astype(dtype)
# Default loss is squared_loss
y[y == 0] = -1
X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.8)
if datatype == "dataframe":
X_train = cudf.DataFrame(X_train)
X_test = cudf.DataFrame(X_test)
y_train = cudf.Series(y_train)
cu_sgd = cumlSGD()
cu_sgd.fit(X_train, y_train)
cu_pred = cu_sgd.predict(X_test)
if datatype == "dataframe":
assert isinstance(cu_pred, cudf.Series)
cu_pred = cu_pred.to_numpy()
else:
assert isinstance(cu_pred, np.ndarray)
# Adjust for squared loss (we don't need to test for high accuracy,
# just that the loss function tended towards the expected classes.
cu_pred[cu_pred < 0] = -1
cu_pred[cu_pred >= 0] = 1
assert np.array_equal(cu_pred, y_test)
def test_svd_default(datatype):
X_train = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]], dtype=datatype)
y_train = np.array([1, 1, 2, 2], dtype=datatype)
X_test = np.array([[3.0, 5.0], [2.0, 5.0]]).astype(datatype)
cu_sgd = cumlSGD()
cu_sgd.fit(X_train, y_train)
cu_pred = cu_sgd.predict(X_test).to_array()
print("cuML predictions : ", cu_pred)
def test_svd(datatype, lrate, input_type, penalty, loss, name):
if name == 'blobs':
n_samples = 500000
train_rows = int(n_samples * 0.8)
X, y = make_blobs(n_samples=n_samples, n_features=1000, random_state=0)
X_test = np.array(X[train_rows:, 0:], dtype=datatype)
X_train = np.array(X[0:train_rows, :], dtype=datatype)
y_train = np.array(y[0:train_rows, ], dtype=datatype)
elif name == 'iris':
iris = datasets.load_iris()
X = iris.data
y = iris.target
train_rows = int((np.shape(X)[0]) * 0.8)
X_test = np.array(X[train_rows:, 0:], dtype=datatype)
X_train = np.array(X[0:train_rows, :], dtype=datatype)
y_train = np.array(y[0:train_rows, ], dtype=datatype)
else:
X_train = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]],
dtype=datatype)
y_train = np.array([1, 1, 2, 2], dtype=datatype)
X_test = np.array([[3.0, 5.0], [2.0, 5.0]]).astype(datatype)
cu_sgd = cumlSGD(learning_rate=lrate,
eta0=0.005,
epochs=2000,
fit_intercept=True,
batch_size=4096,
tol=0.0,
penalty=penalty,
loss=loss)
if input_type == 'dataframe':
y_train_pd = pd.DataFrame({'fea0': y_train[0:, ]})
X_train_pd = pd.DataFrame(
{'fea%d' % i: X_train[0:, i]
for i in range(X_train.shape[1])})
X_test_pd = pd.DataFrame(
{'fea%d' % i: X_test[0:, i]
for i in range(X_test.shape[1])})
X_train = cudf.DataFrame.from_pandas(X_train_pd)
X_test = cudf.DataFrame.from_pandas(X_test_pd)
y_train = y_train_pd.values
y_train = y_train[:, 0]
y_train = cudf.Series(y_train)
cu_sgd.fit(X_train, y_train)
cu_pred = cu_sgd.predict(X_test).to_array()
print("cuML predictions : ", cu_pred)
def test_sgd(dtype, lrate, penalty, loss, datatype):
X, y = make_blobs(n_samples=100, n_features=3, centers=2, random_state=0)
X = X.astype(dtype)
y = y.astype(dtype)
if loss == "hinge" or loss == "squared_loss":
y[y == 0] = -1
X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.8)
if datatype == "dataframe":
X_train = cudf.DataFrame(X_train)
X_test = cudf.DataFrame(X_test)
y_train = cudf.Series(y_train)
cu_sgd = cumlSGD(learning_rate=lrate,
eta0=0.005,
epochs=2000,
fit_intercept=True,
batch_size=4096,
tol=0.0,
penalty=penalty,
loss=loss,
power_t=0.4)
cu_sgd.fit(X_train, y_train)
cu_pred = cu_sgd.predict(X_test)
if datatype == "dataframe":
assert isinstance(cu_pred, cudf.Series)
cu_pred = cu_pred.to_numpy()
else:
assert isinstance(cu_pred, np.ndarray)
if loss == "log":
cu_pred[cu_pred < 0.5] = 0
cu_pred[cu_pred >= 0.5] = 1
elif loss == "squared_loss":
cu_pred[cu_pred < 0] = -1
cu_pred[cu_pred >= 0] = 1
# Adjust for squared loss (we don't need to test for high accuracy,
# just that the loss function tended towards the expected classes.
assert np.array_equal(cu_pred, y_test)
def test_svd(datatype, lrate, input_type, penalty, loss):
X = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]], dtype=datatype)
y = np.array([1, 1, 2, 2], dtype=datatype)
pred_data = np.array([[3.0, 5.0], [2.0, 5.0]]).astype(datatype)
if input_type == 'dataframe':
X = cudf.DataFrame()
X['col1'] = np.asarray([-1, -2, 1, 2], dtype=datatype)
X['col2'] = np.asarray([-1, -1, 2, 2], dtype=datatype)
y = cudf.Series(np.array(y, dtype=np.float32))
pred_data = cudf.DataFrame()
pred_data['col1'] = np.asarray([3, 2], dtype=datatype)
pred_data['col2'] = np.asarray([5, 5], dtype=datatype)
cu_sgd = cumlSGD(learning_rate=lrate, eta0=0.005, epochs=2000,
fit_intercept=True, batch_size=2,
tol=0.0, penalty=penalty, loss=loss)
cu_sgd.fit(X, y)
cu_pred = cu_sgd.predict(pred_data).to_array()
print("cuML predictions : ", cu_pred)
def test_svd(datatype, lrate, penalty, loss, name):
if name == 'blobs':
X, y = make_blobs(n_samples=500000, n_features=1000, random_state=0)
X = X.astype(datatype)
y = y.astype(datatype)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
train_size=0.8)
elif name == 'iris':
iris = datasets.load_iris()
X = (iris.data).astype(datatype)
y = (iris.target).astype(datatype)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
train_size=0.8)
else:
X_train = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]],
dtype=datatype)
y_train = np.array([1, 1, 2, 2], dtype=datatype)
X_test = np.array([[3.0, 5.0], [2.0, 5.0]]).astype(datatype)
cu_sgd = cumlSGD(learning_rate=lrate,
eta0=0.005,
epochs=2000,
fit_intercept=True,
batch_size=4096,
tol=0.0,
penalty=penalty,
loss=loss)
cu_sgd.fit(X_train, y_train)
cu_pred = cu_sgd.predict(X_test).to_array()
print("cuML predictions : ", cu_pred)