def tree_bins(x, y, inf_edges, tree):
X_in = assure_numpy_array(x).reshape(-1, 1)
y_in = assure_numpy_array(y).reshape(-1, 1)
tree.fit(X_in, y_in)
if tree.min_samples_leaf >= X_in.shape[0]:
error_msg = (
"Cannot Fit decision tree. min_samples_leaf must be < than the length of x.m"
+ f"Currently min_samples_leaf {tree.min_samples_leaf} " +
f"and the length of X is {X_in.shape[0]}")
raise ValueError(error_msg)
leaves = tree.apply(X_in)
index, counts = np.unique(leaves, return_counts=True)
bin_count = len(index)
tpf = TreePathFinder(tree)
boundaries = [bound['min'] for bound in tpf.get_boundaries().values()]
boundaries += [tpf.get_boundaries()[leaves[-1]]['max']]
if not inf_edges:
boundaries[0] = np.min(X_in)
boundaries[-1] = np.max(X_in)
return counts, boundaries, bin_count, tree
def tree_bins(x, y, inf_edges, tree):
"""
Tree.
"""
X_in = assure_numpy_array(x).reshape(-1, 1)
y_in = assure_numpy_array(y).reshape(-1, 1)
tree.fit(X_in, y_in)
if tree.min_samples_leaf >= X_in.shape[0]:
error_msg = (
"Cannot Fit decision tree. min_samples_leaf must be < than the length of x.m"
+ f"Currently min_samples_leaf {tree.min_samples_leaf} " +
f"and the length of X is {X_in.shape[0]}")
raise ValueError(error_msg)
leaves = tree.apply(X_in)
index, counts = np.unique(leaves, return_counts=True)
bin_count = len(index)
boundaries = np.unique(
tree.tree_.threshold[tree.tree_.feature != _tree.TREE_UNDEFINED])
boundaries = [np.min(X_in)] + boundaries.tolist() + [np.max(X_in)]
if inf_edges:
boundaries[0] = -np.inf
boundaries[-1] = np.inf
return counts.tolist(), boundaries, bin_count, tree
def test_assure_numpy_array_list():
x = [1, 2, 3]
x_array = assure_numpy_array(x)
assert isinstance(x_array, np.ndarray)
np.testing.assert_array_equal(x_array, np.array(x))
x = [[1, 2], [3, 4]]
x_array = assure_numpy_array(x)
np.testing.assert_array_equal(x_array, np.array([[1, 2], [3, 4]]))
with pytest.raises(DimensionalityError):
assert assure_numpy_array(x, assure_1d=True)
def test_assure_numpy_array_dataframe():
x = pd.DataFrame({'x': [1, 2, 3]})
x_array = assure_numpy_array(x)
assert isinstance(x_array, np.ndarray)
np.testing.assert_array_equal(x_array, np.array([1, 2, 3]))
x = pd.DataFrame({'x': [1, 2, 3], 'y': [1, 2, 3]})
x_array = assure_numpy_array(x)
np.testing.assert_array_equal(x_array, np.array([[1, 1], [2, 2], [3, 3]]))
with pytest.raises(DimensionalityError):
assert assure_numpy_array(x, assure_1d=True)
def test_assure_numpy_array_array():
"""
Test.
"""
x = np.array([1, 2, 3])
x_array = assure_numpy_array(x)
assert isinstance(x_array, np.ndarray)
np.testing.assert_array_equal(x_array, x)
x = np.array([[1, 2], [3, 4]])
x_array = assure_numpy_array(x)
np.testing.assert_array_equal(x_array, x)
with pytest.raises(DimensionalityError):
assert assure_numpy_array(x, assure_1d=True)
def test_assure_numpy_array_series():
"""
Test.
"""
x = pd.Series([1, 2, 3])
x_array = assure_numpy_array(x)
assert isinstance(x_array, np.ndarray)
np.testing.assert_array_equal(x_array, np.array([1, 2, 3]))
def get_metric(X,
y,
clf,
test_size,
split_seed,
scorers,
train_sampling_type=None,
test_sampling_type=None,
train_sampling_fraction=1,
test_sampling_fraction=1):
"""
Draws random train/test sample from the data using random seed and calculates metric of interest.
Args:
X (np.array or pd.DataFrame):
Dataset with features.
y (np.array or pd.Series):
Target of the prediction.
clf (model object):
Binary classification model or pipeline.
test_size (float):
Fraction of data used for testing the model.
split_seed (int):
Randomized seed used for splitting data.
scorers (list of Scorers):
List of Scorer objects used to score the trained model.
train_sampling_type (str, optional):
String indicating what type of sampling should be applied on train set:
- `None`: indicates that no additional sampling is done after splitting data,
- `'bootstrap'`: indicates that sampling with replacement will be performed on train data,
- `'subsample'`: indicates that sampling without repetition will be performed on train data.
test_sampling_type (str, optional):
string indicating what type of sampling should be applied on test set:
- `None`: indicates that no additional sampling is done after splitting data
- `'bootstrap'`: indicates that sampling with replacement will be performed on test data
- `'subsample'`: indicates that sampling without repetition will be performed on test data
train_sampling_fraction (float, optional):
Fraction of train data sampled, if sample_train_type is not None. Default value is 1.
test_sampling_fraction (float, optional):
Fraction of test data sampled, if sample_test_type is not None. Default value is 1.
Returns:
(pd.Dataframe):
Dataframe with results for a given model trained. Rows indicate the metric measured and columns ther results
"""
if not (isinstance(X, np.ndarray) or isinstance(X, pd.DataFrame)):
X = assure_numpy_array(X)
if not (isinstance(X, np.ndarray) or isinstance(X, pd.Series)):
y = assure_numpy_array(y)
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=test_size, random_state=split_seed, stratify=y)
# Sample data based on the input arguments
X_train, y_train = sample_data(X=X_train,
y=y_train,
sampling_type=train_sampling_type,
sampling_fraction=train_sampling_fraction,
dataset_name='train')
X_test, y_test = sample_data(X=X_test,
y=y_test,
sampling_type=test_sampling_type,
sampling_fraction=test_sampling_fraction,
dataset_name='test')
clf = clf.fit(X_train, y_train)
results_columns = [
'metric_name', 'train_score', 'test_score', 'delta_score'
]
results = pd.DataFrame([], columns=results_columns)
for scorer in scorers:
score_train = scorer.score(clf, X_train, y_train)
score_test = scorer.score(clf, X_test, y_test)
score_delta = score_train - score_test
results = results.append(
pd.DataFrame(
[[scorer.metric_name, score_train, score_test, score_delta]],
columns=results_columns))
return results
