def test_two_leaves():
np.random.seed(999)
df = pd.DataFrame([
# x1, x2, y stratify x1, consider y ~ x2
[1, 3, 5],
[2, 4, 6],
[1, 4, 5],
[2, 2, 4]
], columns=['x1','x2','y'])
X = df.drop('y', axis=1)
y = df['y']
leaves = get_leaves(X, y, 'x2') # get index of samples in each leaf
expected_leaves = [np.array([0, 2]), # leaf 0
np.array([1, 3])] # leaf 1
np.testing.assert_array_equal(leaves, expected_leaves)
leaf_deltas, leaf_counts, refcats, ignored = stratify_cats(X,y,colname="x2", min_samples_leaf=2)
expected_leaf_deltas = np.array([[nan, nan], # 0
[nan, nan], # 1
[nan, 0], # 2
[0, nan], # 3
[0, 2]]) # 4
expected_leaf_counts = np.array([[0, 0],
[0, 0],
[0, 1],
[1, 0],
[1, 1]])
expected_refcats = np.array([4, 2])
np.testing.assert_array_almost_equal(leaf_deltas, expected_leaf_deltas, decimal=1)
np.testing.assert_array_equal(leaf_counts, expected_leaf_counts)
np.testing.assert_array_equal(refcats, expected_refcats)
assert ignored==0
def test_single_leaf():
np.random.seed(999)
df = pd.DataFrame([
# x1, x2, y stratify x1, consider y ~ x2
[1, 3, 5],
[1, 4, 6],
[1, 4, 5],
[1, 2, 4]
], columns=['x1','x2','y'])
X = df.drop('y', axis=1)
y = df['y']
leaves = get_leaves(X, y, 'x2') # get index of samples in each leaf
expected_leaves = [np.array([0, 1, 2, 3])] # leaf 0
np.testing.assert_array_equal(leaves, expected_leaves)
leaf_deltas, leaf_counts, refcats, ignored = stratify_cats(X,y,colname="x2", min_samples_leaf=4)
expected_leaf_deltas = np.array([nan, nan, -1, 0, .5]).reshape(-1,1)
expected_leaf_counts = np.array([0, 0, 1, 1, 2]).reshape(-1,1)
expected_refcats = np.array([3])
np.testing.assert_array_almost_equal(leaf_deltas, expected_leaf_deltas, decimal=1)
np.testing.assert_array_equal(leaf_counts, expected_leaf_counts)
np.testing.assert_array_equal(refcats, expected_refcats)
assert ignored==0
def test_three_leaves_no_overlap():
np.random.seed(999)
df = pd.DataFrame(
[
# x1, x2, y stratify x1, consider y ~ x2
[1, 2, 9],
[1, 3, 7],
[3, 4, 6],
[3, 5, 5],
[4, 6, 4],
[4, 7, 3]
],
columns=['x1', 'x2', 'y'])
X = df.drop('y', axis=1)
y = df['y']
leaves = get_leaves(
X, y, 'x2', min_samples_leaf=2) # get index of samples in each leaf
expected_leaves = [
np.array([0, 1]), # leaf 0
np.array([2, 3]), # leaf 1
np.array([4, 5])
] # leaf 2
np.testing.assert_array_equal(leaves, expected_leaves)
leaf_deltas, leaf_counts, ignored = stratify_cats(X,
y,
colname="x2",
min_samples_leaf=2)
print(leaf_deltas, leaf_counts)
expected_leaf_deltas = np.array([
[nan, nan, nan], # cat 0
[nan, nan, nan], # cat 2
[2, nan, nan], # cat 3
[0, nan, nan], # cat 4
[nan, 1, nan],
[nan, 0, nan],
[nan, nan, 1],
[nan, nan, 0]
])
expected_leaf_counts = np.array([[0, 0, 0], [0, 0, 0], [1, 0,
0], [1, 0, 0],
[0, 1, 0], [0, 1, 0], [0, 0, 1],
[0, 0, 1]])
np.testing.assert_array_almost_equal(leaf_deltas,
expected_leaf_deltas,
decimal=1)
np.testing.assert_array_equal(leaf_counts, expected_leaf_counts)
assert ignored == 0
def test_two_leaves_with_2nd_ignored():
np.random.seed(999)
df = pd.DataFrame(
[
# x1, x2, y stratify x1, consider y ~ x2
[1, 3, 5],
[1, 4, 6],
[2, 4, 7],
[2, 4, 8]
],
columns=['x1', 'x2', 'y'])
X = df.drop('y', axis=1)
y = df['y']
"""
Second leaf is indexes 2,3 of x2, which has same x value. must ignore so
there is one leaf, with cats 3 and 4:
leaf_deltas
[[nan]
[nan]
[nan]
[ 0.]
[ 1.]]
"""
leaves = get_leaves(X, y, 'x2') # get index of samples in each leaf
expected_leaves = [
np.array([0, 1]), # leaf 0
np.array([2, 3])
] # leaf 1
np.testing.assert_array_equal(leaves, expected_leaves)
leaf_deltas, leaf_counts, ignored = stratify_cats(X,
y,
colname="x2",
min_samples_leaf=2)
expected_leaf_deltas = np.array([[nan, nan], [nan, nan], [nan, nan],
[0, nan], [1, 0]])
expected_leaf_counts = np.array([[0, 0], [0, 0], [0, 0], [1, 0], [1, 2]])
np.testing.assert_array_almost_equal(leaf_deltas,
expected_leaf_deltas,
decimal=1)
np.testing.assert_array_equal(leaf_counts, expected_leaf_counts)
assert ignored == 0
def speed_ModelID():
"I believe none of this is in the JIT path; repeated runs are same speed"
np.random.seed(1)
n = 20_000
min_samples_leaf = 5
X, y = load_bulldozer(n=n)
leaf_deltas, leaf_counts, ignored = \
stratify_cats(X,y,colname="ModelID",min_samples_leaf=min_samples_leaf)
start = timer()
avg_values_at_cat(leaf_deltas, leaf_counts, max_iter=10)
stop = timer()
nunique = len(np.unique(X['ModelID']))
print(
f"n={n}, unique cats {nunique}, min_samples_leaf={min_samples_leaf}: avg_values_at_cat {stop - start:.3f}s"
)