def test_pipeline(self):
if self.submodule is None:
return
d = sim.LinearSimulation(num_sample=1000,
problem='classification',
num_feature=5)
d.reset()
x_train, y_train, x_test, y_test = d.get()
_, metrics = self.submodule.pipeline(x_train,
y_train,
x_test,
y_test,
problem='classification')
self.assertGreater(metrics['test_acc'], 0.5)
self.assertGreater(metrics['train_acc'], 0.5)
self.assertGreater(metrics['micro_auc'], 0.5)
self.assertGreater(metrics['macro_auc'], 0.5)
d = sim.LinearSimulation(num_sample=1000,
problem='regression',
num_feature=5)
d.reset()
x_train, y_train, x_test, y_test = d.get()
_, metrics = self.submodule.pipeline(x_train,
y_train,
x_test,
y_test,
problem='regression')
self.assertLess(metrics['test_mse'], 1e2)
self.assertLess(metrics['train_mse'], 1e2)
def test_predict(self):
"""Tests the predict() function that this class adds to its parent."""
d = sim.LinearSimulation(num_sample=500, problem='classification')
d.reset()
x_train, y_train, x_test, _ = d.get()
num_class = len(set(y_train))
num_feature = x_train.shape[1]
is_sparse = sparse.issparse(x_train)
clf = dnn.FunctionalKerasClassifier(
build_fn=dnn.keras_build_fn,
num_feature=num_feature,
num_output=num_class,
is_sparse=is_sparse,
verbose=False)
clf.fit(x_train, y_train)
y_pred = clf.predict(x_test)
y_proba = clf.predict_proba(x_test)
# check shape
self.assertEqual(y_pred.shape, (np.size(x_test, 0),))
# check predicted values (should be integer labels)
self.assertTrue(np.all(np.isclose(y_pred, y_pred.astype(int), 0.0001)))
self.assertTrue(np.array_equal(y_pred, np.argmax(y_proba, axis=1)))
def get_sim(dataset,
problem='classification',
which_features='all',
alternate=False,
**kwargs):
"""Get simulated dataset.
Args:
dataset: str dataset name
problem: str type of learning problem; values = 'classification',
'regression'
which_features: str type of features to use; values = 'all', 'inform',
'uninform'
alternate: bool whether alternate experiment is used
**kwargs: Additional args.
Returns:
d: dataset object
API defined in TOOD(jisungkim)
Raises:
ValueError: if dataset is unknown
"""
if dataset == 'sim_sparsity':
return sim.SparsitySimulation(problem=problem,
which_features=which_features,
alternate=alternate,
**kwargs)
if dataset == 'sim_cardinality':
return sim.CardinalitySimulation(problem=problem,
which_features=which_features,
alternate=alternate,
**kwargs)
if dataset == 'sim_linear':
return sim.LinearSimulation(problem=problem, **kwargs)
if dataset == 'sim_multiplicative':
return sim.MultiplicativeSimulation(problem=problem, **kwargs)
if dataset == 'sim_xor':
return sim.XORSimulation(problem=problem, **kwargs)
else:
raise ValueError('Unknown dataset: {}'.format(dataset))
