def test_float_greater_than_one_raises_backward():
"""
Test that float values not on (0, 1) raise.
"""
msg = "^If a float, `n_features` must be on"
with pytest.raises(ValueError, match=msg):
backward(n_features=1.5)
def test_min_features_requirement_backward():
"""
Check that the requirement that at least
two features must be present.
"""
msg = "less than 2 features present."
with pytest.raises(IndexError, match=msg):
backward(X_train=ones((501, 1)), X_val=ones((501, 1)))
def test_n_features_fails_on_string_backward():
"""
Check that backward raises when passed a string
for `n_features`.
"""
msg = "`n_features` must be of type int or float."
with pytest.raises(TypeError, match=msg):
backward(min_change=None, n_features='-0.75')
def test_n_features():
"""
Test various version of n_features
"""
for n_features in (10000, -1.0):
with pytest.raises(ValueError): # should raise
backward(n_features=n_features)
backward(n_features=0.5) # this should not raise
def test_min_change_greater_than_zero_backward():
"""
Check `backward()`'s `min_change` raises when
not greater than zero
"""
msg = "`min_change` must be greater than zero."
with pytest.raises(ValueError, match=msg):
backward(n_features=None, min_change=-0.75)
def test_both_non_none_backward():
"""
Check `backward()` raise when at least one
of `min_change` or `n_features` are not None.
"""
# Note: items in backticks (``) will be in alphabetical order.
msg = "At least one of `min_change` and `n_features` must be None."
with pytest.raises(TypeError, match=msg):
backward(n_features=0.5, min_change=0.3)
def test_selection_class_use_of_criterion():
"""Test Criterion through `forward()` and `backward()."""
msg = "`criterion` must be one of: None, 'aic', 'bic'."
with pytest.raises(ValueError, match=msg):
forward(min_change=0.5, criterion='acc')
with pytest.raises(ValueError, match=msg):
backward(n_features=0.5, criterion='Santa')
def test_too_few_features():
"""
Check that there are enough features for
for selection to be a coherent goal (i.e., >= 2).
"""
X_train = DEFAULT_SELECTION_PARAMS['X_train']
X_train = X_train[:, 0:1]
with pytest.raises(IndexError):
forward(n_features=1, X_train=X_train)
with pytest.raises(IndexError):
backward(n_features=1, X_train=X_train)
def test_loop_exhaust():
"""Text Exhausting backward()'s loop."""
backward(n_features=X_train.shape[-1], min_change=None, _do_not_skip=False)
def test_passing_significant_change():
"""
Test cases where there is a significant `min_change`
during backward selection.
"""
backward(n_features=None, min_change=1, _last_score_punt=True)
def test_bsel_verbose_output():
backward_output = backward(n_features=2, min_change=None, verbose=True)
assert len(backward_output) >= 1
def test_bsel_min_change_output():
backward_output = backward(n_features=None, min_change=10, criterion='aic')
assert len(backward_output) >= 1
def test_bsel_bic_output():
backward_output = backward(n_features=2, min_change=None, criterion='bic')
assert len(backward_output) >= 1
# -----------------------------------------------------------------------------
# Outputs: Run the Forward and Backward Selection Algorithms
# -----------------------------------------------------------------------------
forward_output = forward()
# Run using the other parameter option
forward_output += forward(n_features=1, min_change=None)
# Force the backward selection algorithm to
# select the single feature it thinks is most predictive.
# If implemented correctly, `backward()` should be able to
# identify `true_best_features` as predictive.
backward_output = backward(n_features=1)
# Run using the other parameter option
backward_output += backward(min_change=0.0001, n_features=None)
# -----------------------------------------------------------------------------
# Test outputs: Type
# -----------------------------------------------------------------------------
def output_type(output):
"""
Test that output type is a list.
"""
msg = "Output from the algorithm was not a list."
assert isinstance(output, list), msg