def test_additive_invariance():
"""
Tests additive invariance
i.e. adding constant numbers to X or y array does not change the model coefficients
"""
X, y = random_data_constructor(noise_mag=0.5)
filename = 'testing'
_ = train_linear_model(X, y, filename=filename)
m = load('testing.sav')
coeff_no_additive = float(m.coef_)
X = X + 100
_ = train_linear_model(X, y, filename=filename)
m = load('testing.sav')
coeff_X_additive = float(m.coef_)
y = y - 100
_ = train_linear_model(X, y, filename=filename)
m = load('testing.sav')
coeff_y_additive = float(m.coef_)
# Check that model coefficients for default and additive data are same (very close)
# Note the use of math.isclose function
assert math.isclose(coeff_no_additive, coeff_X_additive, rel_tol=1e-6)
assert math.isclose(coeff_no_additive, coeff_y_additive, rel_tol=1e-6)
def test_noise_impact():
"""
Tests functionality with low and high noise data and expected change in the R^2 score
"""
X, y = random_data_constructor(noise_mag=0.5)
filename = 'testing'
scores_low_noise = train_linear_model(X, y, filename=filename)
X, y = random_data_constructor(noise_mag=5.0)
filename = 'testing'
scores_high_noise = train_linear_model(X, y, filename=filename)
# Check that R^2 scores from high-noise input is less than that of low-noise input
assert scores_high_noise['Train-score'] < scores_low_noise['Train-score']
assert scores_high_noise['Test-score'] < scores_low_noise['Test-score']
def test_model_works_data_range_sign_change():
"""
Tests for functionality with data scaled high and low
"""
# Small-valued data
X, y = fixed_data_constructor()
X = 1.0e-9 * X
y = 1.0e-9 * y
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
# Check that test and train scores are perfectly equal to 1.0
assert scores['Train-score'] == 1.0
assert scores['Test-score'] == 1.0
# Large-valued data
X, y = fixed_data_constructor()
X = 1.0e9 * X
y = 1.0e9 * y
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
# Check that test and train scores are perfectly equal to 1.0
assert scores['Train-score'] == 1.0
assert scores['Test-score'] == 1.0
# X-values are flipped
X, y = fixed_data_constructor()
X = -1 * X
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
# Check that test and train scores are perfectly equal to 1.0
assert scores['Train-score'] == 1.0
assert scores['Test-score'] == 1.0
# y-values are flipped
X, y = fixed_data_constructor()
y = -1 * y
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
# Check that test and train scores are perfectly equal to 1.0
assert scores['Train-score'] == 1.0
assert scores['Test-score'] == 1.0
def test_raised_exception():
"""
Tests for raised exception with pytest.raises context manager
"""
# ValueError
with pytest.raises(ValueError):
# Insert a np.nan into the X array
X, y = random_data_constructor()
X[1] = np.nan
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
# Insert a np.nan into the y array
X, y = random_data_constructor()
y[1] = np.nan
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
with pytest.raises(ValueError) as exception:
# Insert a string into the X array
X, y = random_data_constructor()
X[1] = "A string"
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
assert "could not convert string to float" in str(exception.value)
def test_model_return_object():
"""
Tests the returned object of the modeling function
"""
X, y = random_data_constructor()
scores = train_linear_model(X, y)
#=================================
# TEST SUITE
#=================================
# Check the return object type
assert isinstance(scores, dict)
# Check the length of the returned object
assert len(scores) == 2
# Check the correctness of the names of the returned dict keys
assert 'Train-score' in scores and 'Test-score' in scores
def test_model_save_load():
"""
Tests for the model saving process
"""
X,y = random_data_constructor()
filename = 'testing'
_ = train_linear_model(X,y, filename=filename)
#=================================
# TEST SUITE
#=================================
# Check the model file is created/saved in the directory
assert path.exists('testing.sav')
# Check that the model file can be loaded properly
# (by type checking that it is a sklearn linear regression estimator)
loaded_model = load('testing.sav')
assert isinstance(loaded_model, sklearn.linear_model._base.LinearRegression)
def test_model_return_vals():
"""
Tests for the returned values of the modeling function
"""
X, y = random_data_constructor()
scores = train_linear_model(X, y)
#=================================
# TEST SUITE
#=================================
# Check returned scores' type
assert isinstance(scores['Train-score'], float)
assert isinstance(scores['Test-score'], float)
# Check returned scores' range
assert scores['Train-score'] >= 0.0
assert scores['Train-score'] <= 1.0
assert scores['Test-score'] >= 0.0
assert scores['Test-score'] <= 1.0
def test_wrong_input_raises_assertion():
"""
Tests for various assertion cheks written in the modeling function
"""
X, y = random_data_constructor()
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
#=================================
# TEST SUITE
#=================================
# Test that it handles the case of: X is a string
msg = train_linear_model('X', y)
assert isinstance(msg, AssertionError)
assert msg.args[0] == "X must be a Numpy array"
# Test that it handles the case of: y is a string
msg = train_linear_model(X, 'y')
assert isinstance(msg, AssertionError)
assert msg.args[0] == "y must be a Numpy array"
# Test that it handles the case of: test_frac is a string
msg = train_linear_model(X, y, test_frac='0.2')
assert isinstance(msg, AssertionError)
assert msg.args[0] == "Test set fraction must be a floating point number"
# Test that it handles the case of: test_frac is within 0.0 and 1.0
msg = train_linear_model(X, y, test_frac=-0.2)
assert isinstance(msg, AssertionError)
assert msg.args[0] == "Test set fraction must be between 0.0 and 1.0"
msg = train_linear_model(X, y, test_frac=1.2)
assert isinstance(msg, AssertionError)
assert msg.args[0] == "Test set fraction must be between 0.0 and 1.0"
# Test that it handles the case of: filename for model save a string
msg = train_linear_model(X, y, filename=2.0)
assert isinstance(msg, AssertionError)
assert msg.args[0] == "Filename must be a string"
# Test that function is checking input vector shape compatibility
X = X.reshape(10, 10)
msg = train_linear_model(X, y, filename='testing')
assert isinstance(msg, AssertionError)
assert msg.args[0] == "Row numbers of X and y data must be identical"
def test_loaded_model_works():
"""
Tests if the loading of the model works correctly
"""
X, y = fixed_data_constructor()
if len(X.shape) == 1:
X = X.reshape(-1, 1)
if len(y.shape) == 1:
y = y.reshape(-1, 1)
filename = 'testing'
scores = train_linear_model(X, y, filename=filename)
loaded_model = load('testing.sav')
#=================================
# TEST SUITE
#=================================
# Check that test and train scores are perfectly equal to 1.0
assert scores['Train-score'] == 1.0
assert scores['Test-score'] == 1.0
# Check that trained model predicts the y (almost) perfectly given X
# Note the use of np.testing function instead of standard 'assert'
# To handle numerical precision issues, we should use the `assert_allclose` function instead of any equality check
np.testing.assert_allclose(y, loaded_model.predict(X))
