def test_mbsgd_classifier_default(datatype, nrows, column_info):
ncols, n_info = column_info
X, y = make_classification(n_samples=nrows,
n_informative=n_info,
n_features=ncols,
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,
random_state=0)
y_train = y_train.astype(datatype)
y_test = y_test.astype(datatype)
cu_mbsgd_classifier = cumlMBSGClassifier()
cu_mbsgd_classifier.fit(X_train, y_train)
cu_pred = cu_mbsgd_classifier.predict(X_test).to_array()
skl_sgd_classifier = SGDClassifier()
skl_sgd_classifier.fit(X_train, y_train)
skl_pred = skl_sgd_classifier.predict(X_test)
cu_acc = accuracy_score(cu_pred, y_test)
skl_acc = accuracy_score(skl_pred, y_test)
assert cu_acc >= skl_acc - 0.05
def test_mbsgd_classifier_vs_skl(lrate, penalty, loss, make_dataset):
nrows, X_train, X_test, y_train, y_test = make_dataset
if nrows < 500000:
cu_mbsgd_classifier = cumlMBSGClassifier(learning_rate=lrate,
eta0=0.005,
epochs=100,
fit_intercept=True,
batch_size=2,
tol=0.0,
penalty=penalty)
cu_mbsgd_classifier.fit(X_train, y_train)
cu_pred = cu_mbsgd_classifier.predict(X_test)
cu_acc = accuracy_score(cp.asnumpy(cu_pred), cp.asnumpy(y_test))
skl_sgd_classifier = SGDClassifier(learning_rate=lrate,
eta0=0.005,
max_iter=100,
fit_intercept=True,
tol=0.0,
penalty=penalty,
random_state=0)
skl_sgd_classifier.fit(cp.asnumpy(X_train), cp.asnumpy(y_train))
skl_pred = skl_sgd_classifier.predict(cp.asnumpy(X_test))
skl_acc = accuracy_score(skl_pred, cp.asnumpy(y_test))
assert cu_acc >= skl_acc - 0.08
def test_mbsgd_classifier(datatype, lrate, input_type, penalty,
loss, nrows, column_info):
ncols, n_info = column_info
X, y = make_classification(n_samples=nrows, n_informative=n_info,
n_features=ncols, 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,
random_state=10)
cu_mbsgd_classifier = cumlMBSGClassifier(learning_rate=lrate, eta0=0.005,
epochs=100, fit_intercept=True,
batch_size=2, tol=0.0,
penalty=penalty)
cu_mbsgd_classifier.fit(X_train, y_train)
cu_pred = cu_mbsgd_classifier.predict(X_test).to_array()
cu_acc = accuracy_score(cu_pred, y_test)
if nrows < 500000:
skl_sgd_classifier = SGDClassifier(learning_rate=lrate, eta0=0.005,
max_iter=100, fit_intercept=True,
tol=0.0, penalty=penalty,
random_state=0)
skl_sgd_classifier.fit(X_train, y_train)
skl_pred = skl_sgd_classifier.predict(X_test)
skl_acc = accuracy_score(skl_pred, y_test)
assert cu_acc >= skl_acc - 0.06
def test_mbsgd_classifier_default(make_dataset):
nrows, X_train, X_test, y_train, y_test = make_dataset
cu_mbsgd_classifier = cumlMBSGClassifier(batch_size=nrows / 10)
cu_mbsgd_classifier.fit(X_train, y_train)
cu_pred = cu_mbsgd_classifier.predict(X_test)
cu_acc = accuracy_score(cp.asnumpy(cu_pred), cp.asnumpy(y_test))
assert cu_acc >= 0.69
def test_mbsgd_classifier_set_params():
x = np.linspace(0, 1, 50)
y = (x > 0.5).astype(cp.int32)
model = cumlMBSGClassifier()
model.fit(x, y)
coef_before = model.coef_
model = cumlMBSGClassifier(epochs=20, loss='hinge')
model.fit(x, y)
coef_after = model.coef_
model = cumlMBSGClassifier()
model.set_params(**{'epochs': 20, 'loss': 'hinge'})
model.fit(x, y)
coef_test = model.coef_
assert coef_before != coef_after
assert coef_after == coef_test
def test_mbsgd_classifier_default(make_dataset):
nrows, X_train, X_test, y_train, y_test = make_dataset
cu_mbsgd_classifier = cumlMBSGClassifier()
cu_mbsgd_classifier.fit(X_train, y_train)
cu_pred = cu_mbsgd_classifier.predict(X_test)
cu_acc = accuracy_score(cu_pred, y_test)
assert cu_acc >= 0.69
def test_mbsgd_classifier_attributes():
X, y = make_blobs()
clf = cumlMBSGClassifier()
clf.fit(X, y)
attrs = ["dtype", "solver_model", "coef_", "intercept_",
"l1_ratio", "n_cols", "eta0", "batch_size",
"fit_intercept", "penalty"]
for attr in attrs:
assert hasattr(clf, attr)
def test_mbsgd_classifier(lrate, penalty, loss, make_dataset):
nrows, X_train, X_test, y_train, y_test = make_dataset
cu_mbsgd_classifier = cumlMBSGClassifier(learning_rate=lrate, eta0=0.005,
epochs=100, fit_intercept=True,
batch_size=nrows/100, tol=0.0,
penalty=penalty)
cu_mbsgd_classifier.fit(X_train, y_train)
cu_pred = cu_mbsgd_classifier.predict(X_test)
cu_acc = accuracy_score(cp.asnumpy(cu_pred), cp.asnumpy(y_test))
assert cu_acc > 0.79
def test_mbsgd_classifier_default(make_dataset):
nrows, X_train, X_test, y_train, y_test = make_dataset
cu_mbsgd_classifier = cumlMBSGClassifier()
cu_mbsgd_classifier.fit(X_train, y_train)
cu_pred = cu_mbsgd_classifier.predict(X_test).to_array()
cu_acc = accuracy_score(cu_pred, y_test)
if nrows < 500000:
skl_sgd_classifier = SGDClassifier()
skl_sgd_classifier.fit(X_train, y_train)
skl_pred = skl_sgd_classifier.predict(X_test)
skl_acc = accuracy_score(skl_pred, y_test)
assert cu_acc >= skl_acc - 0.05
def test_mbsgd_classifier(datatype, lrate, input_type, penalty, loss, nrows,
ncols):
train_rows = int(nrows * 0.8)
X, y = make_classification(n_samples=nrows,
n_features=ncols,
random_state=0)
X_test = np.array(X[train_rows:, :], dtype=datatype)
X_train = np.array(X[:train_rows, :], dtype=datatype)
y_train = np.array(y[:train_rows, ], dtype=datatype)
y_test = np.array(y[train_rows:, ], dtype=datatype)
cu_mbsgd_classifier = cumlMBSGClassifier(learning_rate=lrate,
eta0=0.005,
epochs=100,
fit_intercept=True,
batch_size=2,
tol=0.0,
penalty=penalty)
cu_mbsgd_classifier.fit(X_train, y_train)
cu_pred = cu_mbsgd_classifier.predict(X_test).to_array()
skl_sgd_classifier = SGDClassifier(learning_rate=lrate,
eta0=0.005,
max_iter=100,
fit_intercept=True,
tol=0.0,
penalty=penalty,
random_state=0)
skl_sgd_classifier.fit(X_train, y_train)
skl_pred = skl_sgd_classifier.predict(X_test)
cu_error = accuracy_score(cu_pred, y_test)
skl_error = accuracy_score(skl_pred, y_test)
assert (cu_error - skl_error <= 0.02)
