def test_neural_forest_regressor():
"""
This unit test checks the training and evaluating stage of
NeuralForestClassifier.
"""
epochs = 1
n_estimators = 2
depth = 3
lamda = 1e-3
model = NeuralForestRegressor(
n_estimators=n_estimators,
depth=depth,
lamda=lamda,
cuda=False,
n_jobs=1,
)
model.set_optimizer("Adam", lr=1e-4, weight_decay=5e-4)
# Prepare data
train = TensorDataset(X_train, y_train_reg)
train_loader = DataLoader(train, batch_size=2, shuffle=False)
test = TensorDataset(X_test, y_test_reg)
test_loader = DataLoader(test, batch_size=2, shuffle=False)
# Train
model.fit(train_loader, epochs=epochs, test_loader=test_loader)
# Evaluate
model.evaluate(test_loader)
# Predict
for _, (data, target) in enumerate(test_loader):
model.predict(data)
break
# Reload
new_model = NeuralForestRegressor(
n_estimators=n_estimators,
depth=depth,
lamda=lamda,
cuda=False,
n_jobs=1,
)
io.load(new_model)
new_model.evaluate(test_loader)
for _, (data, target) in enumerate(test_loader):
new_model.predict(data)
break
def test_clf_class(clf):
"""
This unit test checks the training and evaluating stage of all classifiers.
"""
epochs = 1
n_estimators = 2
model = clf(estimator=MLP_clf, n_estimators=n_estimators, cuda=False)
# Optimizer
model.set_optimizer("Adam", lr=1e-3, weight_decay=5e-4)
# Scheduler (Snapshot Ensemble Excluded)
if not isinstance(model, torchensemble.SnapshotEnsembleClassifier):
model.set_scheduler("MultiStepLR", milestones=[2, 4])
# Prepare data
train = TensorDataset(X_train, y_train_clf)
train_loader = DataLoader(train, batch_size=2, shuffle=False)
test = TensorDataset(X_test, y_test_clf)
test_loader = DataLoader(test, batch_size=2, shuffle=False)
# Snapshot Ensemble needs more epochs
if isinstance(model, torchensemble.SnapshotEnsembleClassifier):
epochs = 6
# Train
ret = model.fit(train_loader, epochs=epochs, test_loader=test_loader)
# Extra step for Fast Geometric Ensemble
if isinstance(model, torchensemble.FastGeometricClassifier):
model.ensemble(ret, train_loader, test_loader=test_loader)
# Evaluate
model.evaluate(test_loader)
# Predict
for _, (data, target) in enumerate(test_loader):
model.predict(data)
break
# Reload
new_model = clf(estimator=MLP_clf, n_estimators=n_estimators, cuda=False)
io.load(new_model)
new_model.evaluate(test_loader)
for _, (data, target) in enumerate(test_loader):
new_model.predict(data)
break
def test_reg_object(reg):
"""
This unit test checks the training and evaluating stage of all regressors.
"""
epochs = 1
n_estimators = 2
model = reg(estimator=MLP_reg(), n_estimators=n_estimators, cuda=False)
# Optimizer
model.set_optimizer("Adam", lr=1e-3, weight_decay=5e-4)
# Scheduler (Snapshot Ensemble Excluded)
if not isinstance(model, torchensemble.SnapshotEnsembleRegressor):
model.set_scheduler("MultiStepLR", milestones=[2, 4])
# Prepare data with multiple inputs
train = TensorDataset(X_train, X_train, y_train_reg)
train_loader = DataLoader(train, batch_size=2, shuffle=False)
test = TensorDataset(X_test, X_test, y_test_reg)
test_loader = DataLoader(test, batch_size=2, shuffle=False)
# Snapshot Ensemble needs more epochs
if isinstance(model, torchensemble.SnapshotEnsembleRegressor):
epochs = 6
# Train
model.fit(train_loader, epochs=epochs, test_loader=test_loader)
# Evaluate
model.evaluate(test_loader)
# Predict
for _, elem in enumerate(test_loader):
data, target = io.split_data_target(elem, device)
model.predict(*data)
break
# Reload
new_model = reg(estimator=MLP_reg(), n_estimators=n_estimators, cuda=False)
io.load(new_model)
new_model.evaluate(test_loader)
for _, elem in enumerate(test_loader):
data, target = io.split_data_target(elem, device)
new_model.predict(*data)
break
def test_clf_class(clf):
"""
This unit test checks the training and evaluating stage of all classifiers.
"""
epochs = 1
n_estimators = 2
model = clf(estimator=MLP_clf, n_estimators=n_estimators, cuda=False)
# Optimizer
model.set_optimizer("Adam", lr=1e-3, weight_decay=5e-4)
# Scheduler (Snapshot Ensemble Excluded)
if not isinstance(model, torchensemble.SnapshotEnsembleClassifier):
model.set_scheduler("MultiStepLR", milestones=[2, 4])
# Prepare data
train = TensorDataset(X_train, y_train_clf)
train_loader = DataLoader(train, batch_size=2, shuffle=False)
test = TensorDataset(X_test, y_test_clf)
test_loader = DataLoader(test, batch_size=2, shuffle=False)
# Snapshot ensemble needs more epochs
if isinstance(model, torchensemble.SnapshotEnsembleClassifier):
epochs = 6
# Train
model.fit(train_loader,
epochs=epochs,
test_loader=test_loader,
save_model=True)
# Test
prev_acc = model.predict(test_loader)
# Reload
new_model = clf(estimator=MLP_clf, n_estimators=n_estimators, cuda=False)
io.load(new_model)
post_acc = new_model.predict(test_loader)
assert prev_acc == post_acc # ensure the same performance
def test_reg_object(reg):
"""
This unit test checks the training and evaluating stage of all regressors.
"""
epochs = 1
n_estimators = 2
model = reg(estimator=MLP_reg(), n_estimators=n_estimators, cuda=False)
# Optimizer
model.set_optimizer("Adam", lr=1e-3, weight_decay=5e-4)
# Scheduler (Snapshot Ensemble Excluded)
if not isinstance(model, torchensemble.SnapshotEnsembleRegressor):
model.set_scheduler("MultiStepLR", milestones=[2, 4])
# Prepare data
train = TensorDataset(X_train, y_train_reg)
train_loader = DataLoader(train, batch_size=2, shuffle=False)
test = TensorDataset(X_test, y_test_reg)
test_loader = DataLoader(test, batch_size=2, shuffle=False)
# Snapshot ensemble needs more epochs
if isinstance(model, torchensemble.SnapshotEnsembleRegressor):
epochs = 6
# Train
ret = model.fit(train_loader, epochs=epochs, test_loader=test_loader)
# Extra step for Fast Geometric Ensemble
if isinstance(model, torchensemble.FastGeometricRegressor):
model.ensemble(ret, train_loader, test_loader=test_loader)
# Test
model.predict(test_loader)
# Reload
new_model = reg(estimator=MLP_reg(), n_estimators=n_estimators, cuda=False)
io.load(new_model)
new_model.predict(test_loader)
