def test_dask_vs_numpy():
client = Client()
ddf = dd.read_csv("tests/test_attributes.csv")
ddf = ddf.repartition(npartitions=4)
g_ddf = gam.GAM(attributions=ddf)
g_ddf.generate()
df = pd.read_csv("tests/test_attributes.csv")
g_df = gam.GAM(attributions=df)
g_df.generate()
assert g_ddf.attributions.all() == g_df.attributions.all()
assert g_ddf.feature_labels == g_df.feature_labels
client.close()
def test_find_optimal_clusters():
""""Create sample attributions with 4 clusters"""
attributions_file = "test_opt.csv"
X, y = make_blobs(
n_samples=40,
n_features=2,
centers=4,
cluster_std=0.01,
center_box=(0.0, 1.0),
shuffle=True,
random_state=42,
)
logging.info(f"blobs made - {X.shape},{y.shape}")
df = pd.DataFrame(columns=["x1", "x2"], data=X)
df.to_csv(attributions_file)
""""Check cluster search via silhouette score"""
g = gam.GAM(attributions_path=attributions_file, distance="kendall_tau")
g.get_optimal_clustering(max_clusters=6, verbose=True)
logging.info(f"attributions file - {g.attributions_path}")
logging.info(f"what we settled on - {g.k}")
assert g.silh_scores == [
(-0.6997008553383604, 4),
(-0.49646173501281243, 3),
(-0.20264778322315316, 6),
(-0.15419026951164008, 5),
(0.3760497014154821, 2),
]
assert g.k == 2
def test_read_csv():
g = gam.GAM(attributions_path="tests/test_attributes.csv")
g._read_local()
assert hasattr(g, "attributions")
assert g.attributions.shape == (4, 3)
assert hasattr(g, "feature_labels")
assert g.feature_labels == ["a1", "a2", "a3"]
def test_read_csv():
g = gam.GAM(attributions_path="tests/test_attributes.csv")
g._read_local()
assert (hasattr(g, 'attributions'))
assert (g.attributions.shape == (4, 3))
assert (hasattr(g, 'feature_labels'))
assert (g.feature_labels == ['a1', 'a2', 'a3'])
def test_normalize():
"""Tests normalization of attributions from csv"""
g = gam.GAM(attributions_path="tests/test_attributes.csv")
g._read_local()
normalized_attributions = gam.GAM.normalize(g.attributions)
assert normalized_attributions.shape == g.attributions.shape
assert not np.any(np.where(normalized_attributions < 0))
assert normalized_attributions.sum(axis=1)[0] == pytest.approx(1.0)
def test_plotting_2attributes():
explanations = [[('height', .05), ('weight', .05), ('hair color', .9)],
[('height', .9), ('weight', .05), ('hair color', .05)]]
g = gam.GAM(attributions_path="tests/test_attributes.csv",
k=len(explanations))
g.explanations = explanations
fname = 'tests/image3'
g.plot(num_features=2, output_path_base=fname, display=False)
output = glob.glob(fname + '*')
assert (len(output) > 0)
for ofile in output:
os.remove(ofile)
def test_plotting_2attributes():
explanations = [
[("height", 0.05), ("weight", 0.05), ("hair color", 0.9)],
[("height", 0.9), ("weight", 0.05), ("hair color", 0.05)],
]
g = gam.GAM(attributions_path="tests/test_attributes.csv", k=len(explanations))
g.explanations = explanations
fname = "tests/image3"
g.plot(num_features=2, output_path_base=fname, display=False)
output = glob.glob(fname + "*")
assert len(output) > 0
for ofile in output:
os.remove(ofile)
def test_cluster():
"""Tests subpopulations generated by clustering attributions"""
g = gam.GAM(attributions_path="tests/test_attributes.csv")
g._read_local()
# g.normalized_attributions = gam.GAM.normalize(g.attributions)
g.clustering_attributions = gam.GAM.normalize(g.attributions)
g._cluster()
assert len(g.explanations) == 2
assert g.subpopulation_sizes[0] > 0
assert g.subpopulation_sizes[1] > 0
assert len(g.explanations) == 2
assert g.explanations[0][0][0] == g.feature_labels[0]
first_explanation_sum = sum([weight for label, weight in g.explanations[0]])
assert first_explanation_sum == pytest.approx(1)
second_explanation_sum = sum([weight for label, weight in g.explanations[1]])
assert second_explanation_sum == pytest.approx(1)
def test_find_optimal_clusters():
""""Create sample attributions with 4 clusters"""
attributions_file = 'test_opt.csv'
X, y = make_blobs(n_samples=40, n_features=2, centers=4,
cluster_std=0.01, center_box=(0.0, 1.0),
shuffle=True, random_state=42)
print('blobs made - ', X.shape, y.shape)
df = pd.DataFrame(columns=['x1', 'x2'], data=X)
df.to_csv(attributions_file)
""""Check cluster search via silhouette score"""
g = gam.GAM(attributions_path=attributions_file, distance="kendall_tau")
g.get_optimal_clustering()
print('attributions file - ', g.attributions_path)
print('data size = ', g.normalized_attributions.shape)
print('what we settled on - ', g.k)
assert(g.k == 4)
def test_read_df_or_list():
# preprocessing
df = pd.read_csv("tests/test_attributes.csv")
att_list = df.values.tolist()
feat_labels_list = df.columns.tolist()
att_arr = np.asarray(df.values.tolist())
feat_labels_arr = np.asarray(df.columns.tolist())
client = Client()
ddf = dd.read_csv("tests/test_attributes.csv")
ddf = ddf.repartition(npartitions=4)
dask_att_arr = da.from_array(att_list)
dask_feat_labels_arr = da.from_array(feat_labels_list)
# Testing dask DataFrame
g_ddf = gam.GAM(attributions=ddf)
g_ddf.generate()
assert hasattr(g_ddf, "attributions")
assert g_ddf.attributions.shape == (4, 3)
assert hasattr(g_ddf, "feature_labels")
assert g_ddf.feature_labels == ["a1", "a2", "a3"]
# Testing dask array
g_dask_list = gam.GAM(attributions=dask_att_arr, batchsize=100, feature_labels=dask_feat_labels_arr)
g_dask_list.generate()
assert hasattr(g_dask_list, "attributions")
assert g_dask_list.attributions.shape == (4, 3)
assert hasattr(g_dask_list, "feature_labels")
assert g_dask_list.feature_labels == ["a1", "a2", "a3"]
client.close()
# Testing DataFrame
g_df = gam.GAM(attributions=df)
g_df.generate()
assert hasattr(g_df, "attributions")
assert g_df.attributions.shape == (4, 3)
assert hasattr(g_df, "feature_labels")
assert g_df.feature_labels == ["a1", "a2", "a3"]
# Testing lists
g_list = gam.GAM(attributions=att_list, batchsize=100, feature_labels=feat_labels_list)
g_list.generate()
assert hasattr(g_list, "attributions")
assert g_list.attributions.shape == (4, 3)
assert hasattr(g_list, "feature_labels")
assert g_list.feature_labels == ["a1", "a2", "a3"]
# Testing numpy arrays
g_arr = gam.GAM(attributions=att_arr, batchsize=100, feature_labels=feat_labels_arr)
g_arr.generate()
assert hasattr(g_arr, "attributions")
assert g_arr.attributions.shape == (4, 3)
assert hasattr(g_arr, "feature_labels")
assert g_arr.feature_labels == ["a1", "a2", "a3"]
# Testing failure
with pytest.raises(ValueError):
g_fail = gam.GAM(attributions=att_arr, batchsize=100)
g_fail.generate()
