test_umap_ops.py

# ===================================================
#  UMAP Fit and Transform Operations Test cases
#  (not really fitting anywhere else)
# ===================================================

from sklearn.datasets import make_blobs
from sklearn.cluster import KMeans
from sklearn.metrics import adjusted_rand_score, pairwise_distances
from sklearn.preprocessing import normalize
from numpy.testing import assert_array_equal
from umap import UMAP
from umap.spectral import component_layout
import numpy as np
import scipy.sparse
import pytest
import warnings
from umap.distances import pairwise_special_metric
from umap.utils import disconnected_vertices
from scipy.sparse import csr_matrix

# Transform isn't stable under batching; hard to opt out of this.
# @SkipTest
# def test_scikit_learn_compatibility():
#     check_estimator(UMAP)


# This test is currently to expensive to run when turning
# off numba JITting to detect coverage.
# @SkipTest
# def test_umap_regression_supervision(): # pragma: no cover
#     boston = load_boston()
#     data = boston.data
#     embedding = UMAP(n_neighbors=10,
#                      min_dist=0.01,
#                      target_metric='euclidean',
#                      random_state=42).fit_transform(data, boston.target)
#


# Umap Clusterability
def test_blobs_cluster():
    data, labels = make_blobs(n_samples=500, n_features=10, centers=5)
    embedding = UMAP(n_epochs=100).fit_transform(data)
    assert adjusted_rand_score(labels, KMeans(5).fit_predict(embedding)) == 1.0


# Multi-components Layout
def test_multi_component_layout():
    data, labels = make_blobs(
        100, 2, centers=5, cluster_std=0.5, center_box=[-20, 20], random_state=42
    )

    true_centroids = np.empty((labels.max() + 1, data.shape[1]), dtype=np.float64)

    for label in range(labels.max() + 1):
        true_centroids[label] = data[labels == label].mean(axis=0)

    true_centroids = normalize(true_centroids, norm="l2")

    embedding = UMAP(n_neighbors=4, n_epochs=100).fit_transform(data)
    embed_centroids = np.empty((labels.max() + 1, data.shape[1]), dtype=np.float64)
    embed_labels = KMeans(n_clusters=5).fit_predict(embedding)

    for label in range(embed_labels.max() + 1):
        embed_centroids[label] = data[embed_labels == label].mean(axis=0)

    embed_centroids = normalize(embed_centroids, norm="l2")

    error = np.sum((true_centroids - embed_centroids) ** 2)

    assert error < 15.0, "Multi component embedding to far astray"


# Multi-components Layout
def test_multi_component_layout_precomputed():
    data, labels = make_blobs(
        100, 2, centers=5, cluster_std=0.5, center_box=[-20, 20], random_state=42
    )
    dmat = pairwise_distances(data)

    true_centroids = np.empty((labels.max() + 1, data.shape[1]), dtype=np.float64)

    for label in range(labels.max() + 1):
        true_centroids[label] = data[labels == label].mean(axis=0)

    true_centroids = normalize(true_centroids, norm="l2")

    embedding = UMAP(n_neighbors=4, metric="precomputed", n_epochs=100).fit_transform(
        dmat
    )
    embed_centroids = np.empty((labels.max() + 1, data.shape[1]), dtype=np.float64)
    embed_labels = KMeans(n_clusters=5).fit_predict(embedding)

    for label in range(embed_labels.max() + 1):
        embed_centroids[label] = data[embed_labels == label].mean(axis=0)

    embed_centroids = normalize(embed_centroids, norm="l2")

    error = np.sum((true_centroids - embed_centroids) ** 2)

    assert error < 15.0, "Multi component embedding to far astray"


@pytest.mark.parametrize("num_isolates", [1, 5])
@pytest.mark.parametrize("metric", ["jaccard", "hellinger"])
@pytest.mark.parametrize("force_approximation", [True, False])
def test_disconnected_data(num_isolates, metric, force_approximation):
    options = [False, True]
    disconnected_data = np.random.choice(a=options, size=(10, 30), p=[0.6, 1 - 0.6])
    # Add some disconnected data for the corner case test
    disconnected_data = np.vstack(
        [disconnected_data, np.zeros((num_isolates, 30), dtype="bool")]
    )
    new_columns = np.zeros((num_isolates + 10, num_isolates), dtype="bool")
    for i in range(num_isolates):
        new_columns[10 + i, i] = True
    disconnected_data = np.hstack([disconnected_data, new_columns])

    with pytest.warns(None) as w:
        model = UMAP(
            n_neighbors=3,
            metric=metric,
            force_approximation_algorithm=force_approximation,
        ).fit(disconnected_data)
    assert len(w) >= 1  # at least one warning should be raised here
    # we can't guarantee the order that the warnings will be raised in so check them all.
    flag = 0
    if num_isolates == 1:
        warning_contains = "A few of your vertices"
    elif num_isolates > 1:
        warning_contains = "A large number of your vertices"
    for wn in w:
        flag += warning_contains in str(wn.message)

    isolated_vertices = disconnected_vertices(model)
    assert flag == 1, str(([wn.message for wn in w], isolated_vertices))
    # Check that the first isolate has no edges in our umap.graph_
    assert isolated_vertices[10] == True
    number_of_nan = np.sum(np.isnan(model.embedding_[isolated_vertices]))
    assert number_of_nan >= num_isolates * model.n_components


@pytest.mark.parametrize("num_isolates", [1])
@pytest.mark.parametrize("sparse", [True, False])
def test_disconnected_data_precomputed(num_isolates, sparse):
    disconnected_data = np.random.choice(
        a=[False, True], size=(10, 20), p=[0.66, 1 - 0.66]
    )
    # Add some disconnected data for the corner case test
    disconnected_data = np.vstack(
        [disconnected_data, np.zeros((num_isolates, 20), dtype="bool")]
    )
    new_columns = np.zeros((num_isolates + 10, num_isolates), dtype="bool")
    for i in range(num_isolates):
        new_columns[10 + i, i] = True
    disconnected_data = np.hstack([disconnected_data, new_columns])
    dmat = pairwise_special_metric(disconnected_data)
    if sparse:
        dmat = csr_matrix(dmat)
    model = UMAP(n_neighbors=3, metric="precomputed", disconnection_distance=1).fit(
        dmat
    )

    # Check that the first isolate has no edges in our umap.graph_
    isolated_vertices = disconnected_vertices(model)
    assert isolated_vertices[10] == True
    number_of_nan = np.sum(np.isnan(model.embedding_[isolated_vertices]))
    assert number_of_nan >= num_isolates * model.n_components


# ---------------
# Umap Transform
# --------------


def test_bad_transform_data(nn_data):
    u = UMAP().fit([[1, 1, 1, 1]])
    with pytest.raises(ValueError):
        u.transform([[0, 0, 0, 0]])


# Transform Stability
# -------------------
def test_umap_transform_embedding_stability(iris, iris_subset_model, iris_selection):
    """Test that transforming data does not alter the learned embeddings

    Issue #217 describes how using transform to embed new data using a
    trained UMAP transformer causes the fitting embedding matrix to change
    in cases when the new data has the same number of rows as the original
    training data.
    """

    data = iris.data[iris_selection]
    fitter = iris_subset_model
    original_embedding = fitter.embedding_.copy()

    # The important point is that the new data has the same number of rows
    # as the original fit data
    new_data = np.random.random(data.shape)
    _ = fitter.transform(new_data)

    assert_array_equal(
        original_embedding,
        fitter.embedding_,
        "Transforming new data changed the original embeddings",
    )

    # Example from issue #217
    a = np.random.random((100, 10))
    b = np.random.random((100, 5))

    umap = UMAP(n_epochs=100)
    u1 = umap.fit_transform(a[:, :5])
    u1_orig = u1.copy()
    assert_array_equal(u1_orig, umap.embedding_)

    _ = umap.transform(b)
    assert_array_equal(u1_orig, umap.embedding_)


# -----------
# UMAP Update
# -----------
def test_umap_update(iris, iris_subset_model, iris_selection, iris_model):

    new_data = iris.data[~iris_selection]
    new_model = iris_subset_model
    new_model.update(new_data)

    comparison_graph = scipy.sparse.vstack(
        [iris_model.graph_[iris_selection], iris_model.graph_[~iris_selection]]
    )
    comparison_graph = scipy.sparse.hstack(
        [comparison_graph[:, iris_selection], comparison_graph[:, ~iris_selection]]
    )

    error = np.sum(np.abs((new_model.graph_ - comparison_graph).data))

    assert error < 1.0


def test_umap_update_large(
    iris, iris_subset_model_large, iris_selection, iris_model_large
):

    new_data = iris.data[~iris_selection]
    new_model = iris_subset_model_large
    new_model.update(new_data)

    comparison_graph = scipy.sparse.vstack(
        [
            iris_model_large.graph_[iris_selection],
            iris_model_large.graph_[~iris_selection],
        ]
    )
    comparison_graph = scipy.sparse.hstack(
        [comparison_graph[:, iris_selection], comparison_graph[:, ~iris_selection]]
    )

    error = np.sum(np.abs((new_model.graph_ - comparison_graph).data))

    assert error < 3.0 # Higher error tolerance based on approx nearest neighbors


# -----------------
# UMAP Graph output
# -----------------
def test_umap_graph_layout():
    data, labels = make_blobs(n_samples=500, n_features=10, centers=5)
    model = UMAP(n_epochs=100, transform_mode="graph")
    graph = model.fit_transform(data)
    assert scipy.sparse.issparse(graph)
    nc, cl = scipy.sparse.csgraph.connected_components(graph)
    assert nc == 5

    new_graph = model.transform(data[:10] + np.random.normal(0.0, 0.1, size=(10, 10)))
    assert scipy.sparse.issparse(graph)
    assert new_graph.shape[0] == 10


# ------------------------
# Component layout options
# ------------------------


def test_component_layout_options(nn_data):
    dmat = pairwise_distances(nn_data[:1000])
    n_components = 5
    component_labels = np.repeat(np.arange(5), dmat.shape[0] // 5)
    single = component_layout(
        dmat,
        n_components,
        component_labels,
        2,
        np.random,
        metric="precomputed",
        metric_kwds={"linkage": "single"},
    )
    average = component_layout(
        dmat,
        n_components,
        component_labels,
        2,
        np.random,
        metric="precomputed",
        metric_kwds={"linkage": "average"},
    )
    complete = component_layout(
        dmat,
        n_components,
        component_labels,
        2,
        np.random,
        metric="precomputed",
        metric_kwds={"linkage": "complete"},
    )

    assert single.shape[0] == 5
    assert average.shape[0] == 5
    assert complete.shape[0] == 5

    assert not np.all(single == average)
    assert not np.all(single == complete)
    assert not np.all(average == complete)