def test_mtry(self, causal_X, causal_y, causal_w, mtry):
forest = GRFForestInstrumentalRegressor(mtry=mtry)
forest.fit(causal_X, causal_y, causal_w, causal_w)
if mtry is not None:
assert forest.mtry_ == mtry
else:
assert forest.mtry_ == 5
def test_fit(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor()
with pytest.raises(NotFittedError):
check_is_fitted(forest)
forest.fit(causal_X, causal_y, causal_w, causal_w)
check_is_fitted(forest)
assert hasattr(forest, "grf_forest_")
assert hasattr(forest, "mtry_")
assert forest.criterion == "mse"
def test_with_X_nan(self, causal_X, causal_y, causal_w):
causal_X_nan = causal_X.copy()
index = np.random.choice(causal_X_nan.size, 100, replace=False)
causal_X_nan.ravel()[index] = np.nan
assert np.sum(np.isnan(causal_X_nan)) == 100
forest = GRFForestInstrumentalRegressor()
forest.fit(causal_X_nan, causal_y, causal_w, causal_w)
pred = forest.predict(causal_X_nan)
assert len(pred) == causal_X_nan.shape[0]
def test_equalize_cluster_weights(
self,
causal_X,
causal_y,
causal_w,
causal_cluster,
equalize_cluster_weights,
):
forest = GRFForestInstrumentalRegressor(
equalize_cluster_weights=equalize_cluster_weights)
forest.fit(causal_X,
causal_y,
causal_w,
causal_w,
cluster=causal_cluster)
if equalize_cluster_weights:
assert forest.samples_per_cluster_ == 20
else:
assert forest.samples_per_cluster_ == causal_y.shape[0] - 20
if equalize_cluster_weights:
with pytest.raises(ValueError):
forest.fit(
causal_X,
causal_y,
causal_w,
causal_w,
cluster=causal_cluster,
sample_weight=causal_y,
)
forest.fit(causal_X, causal_y, causal_w, causal_w, cluster=None)
assert forest.samples_per_cluster_ == 0
def test_serialize(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor()
# not fitted
tf = tempfile.TemporaryFile()
pickle.dump(forest, tf)
tf.seek(0)
forest = pickle.load(tf)
forest.fit(causal_X, causal_y, causal_w, causal_w)
# fitted
tf = tempfile.TemporaryFile()
pickle.dump(forest, tf)
tf.seek(0)
new_forest = pickle.load(tf)
pred = new_forest.predict(causal_X)
assert len(pred) == causal_X.shape[0]
def test_estimators_(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor(n_estimators=10)
with pytest.raises(AttributeError):
_ = forest.estimators_
forest.fit(causal_X, causal_y, causal_w, causal_w)
with pytest.raises(ValueError):
_ = forest.estimators_
forest = GRFForestInstrumentalRegressor(n_estimators=10,
enable_tree_details=True)
forest.fit(causal_X, causal_y, causal_w, causal_w)
estimators = forest.estimators_
assert len(estimators) == 10
assert isinstance(estimators[0], GRFTreeInstrumentalRegressor)
check_is_fitted(estimators[0])
def test_alpha(self, causal_X, causal_y, causal_w, alpha):
forest = GRFForestInstrumentalRegressor(alpha=alpha)
if alpha <= 0 or alpha >= 0.25:
with pytest.raises(ValueError):
forest.fit(causal_X, causal_y, causal_w, causal_w)
else:
forest.fit(causal_X, causal_y, causal_w, causal_w)
def test_get_kernel_weights(self, causal_X, causal_y, causal_w):
X_train, X_test, y_train, y_test, w_train, w_test = train_test_split(
causal_X, causal_y, causal_w, test_size=0.33, random_state=42)
forest = GRFForestInstrumentalRegressor()
forest.fit(X_train, y_train, w_train, w_train)
weights = forest.get_kernel_weights(X_test)
assert weights.shape[0] == X_test.shape[0]
assert weights.shape[1] == X_train.shape[0]
oob_weights = forest.get_kernel_weights(X_train, True)
assert oob_weights.shape[0] == X_train.shape[0]
assert oob_weights.shape[1] == X_train.shape[0]
def test_honesty_fraction(self, causal_X, causal_y, causal_w,
honesty_fraction):
forest = GRFForestInstrumentalRegressor(
honesty=True,
honesty_fraction=honesty_fraction,
honesty_prune_leaves=True)
if honesty_fraction <= 0 or honesty_fraction >= 1:
with pytest.raises(RuntimeError):
forest.fit(causal_X, causal_y, causal_w, causal_w)
else:
forest.fit(causal_X, causal_y, causal_w, causal_w)
def test_clone(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor()
forest.fit(causal_X, causal_y, causal_w, causal_w)
clone(forest)
def test_predict(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor()
forest.fit(causal_X, causal_y, causal_w, causal_w)
pred = forest.predict(causal_X)
assert len(pred) == causal_X.shape[0]
def test_get_feature_importances(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor()
forest.fit(causal_X, causal_y, causal_w, causal_w)
fi = forest.get_feature_importances()
assert len(fi) == causal_X.shape[1]
def test_honesty(self, causal_X, causal_y, causal_w, honesty):
forest = GRFForestInstrumentalRegressor(honesty=honesty)
forest.fit(causal_X, causal_y, causal_w, causal_w)
def test_sample_fraction(self, causal_X, causal_y, causal_w,
sample_fraction): # and ci_group_size
forest = GRFForestInstrumentalRegressor(
sample_fraction=sample_fraction, ci_group_size=1)
if sample_fraction <= 0 or sample_fraction > 1:
with pytest.raises(ValueError):
forest.fit(causal_X, causal_y, causal_w, causal_w)
else:
forest.fit(causal_X, causal_y, causal_w, causal_w)
forest = GRFForestInstrumentalRegressor(
sample_fraction=sample_fraction, ci_group_size=2)
if sample_fraction <= 0 or sample_fraction > 0.5:
with pytest.raises(ValueError):
forest.fit(causal_X, causal_y, causal_w, causal_w)
else:
forest.fit(causal_X, causal_y, causal_w, causal_w)
def test_get_estimator(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor(n_estimators=10)
with pytest.raises(NotFittedError):
_ = forest.get_estimator(idx=0)
forest.fit(causal_X, causal_y, causal_w, causal_w)
with pytest.raises(ValueError):
_ = forest.get_estimator(idx=0)
forest = GRFForestInstrumentalRegressor(n_estimators=10,
enable_tree_details=True)
forest.fit(causal_X, causal_y, causal_w, causal_w)
estimator = forest.get_estimator(0)
check_is_fitted(estimator)
assert isinstance(estimator, GRFTreeInstrumentalRegressor)
with pytest.raises(IndexError):
_ = forest.get_estimator(idx=20)
def test_init(self):
_ = GRFForestInstrumentalRegressor()
def test_honesty_prune_leaves(self, causal_X, causal_y, causal_w,
honesty_prune_leaves):
forest = GRFForestInstrumentalRegressor(
honesty=True, honesty_prune_leaves=honesty_prune_leaves)
forest.fit(causal_X, causal_y, causal_w, causal_w)
def test_from_forest(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor()
forest.fit(causal_X, causal_y, causal_w, causal_w)
tree = GRFTreeInstrumentalRegressor.from_forest(forest=forest, idx=0)
tree.predict(causal_X)
def test_get_split_frequencies(self, causal_X, causal_y, causal_w):
forest = GRFForestInstrumentalRegressor()
forest.fit(causal_X, causal_y, causal_w, causal_w)
sf = forest.get_split_frequencies()
assert sf.shape[1] == causal_X.shape[1]