def test_upsample():
data, labels = make_blobs(n_samples=3000,
n_features=2,
centers=3,
random_state=42)
data = pd.DataFrame(data, columns=["X", "Y"])
g = gate.Gate(gate_name="test",
parent="test parent",
x="X",
y="Y",
method="manual",
sampling={
"method": "uniform",
"frac": 0.5
})
sample = g._downsample(data=data)
sample_labels = labels[sample.index.values]
pops = list()
for x in np.unique(sample_labels):
idx = sample.index.values[np.where(sample_labels == x)[0]]
pops.append(
gate.Population(population_name=f"Pop_{x}",
parent="root",
index=idx[:498]))
pops = g._upsample(data=data, sample=sample, populations=pops)
assert isinstance(pops, list)
assert all([isinstance(p, gate.Population) for p in pops])
assert all([len(p.index) == 1000 for p in pops])
for x in np.unique(labels):
p = [i for i in pops if i.population_name == f"Pop_{x}"][0]
assert np.array_equal(p.index, np.where(labels == x)[0])
def test_downsample(kwargs):
g = gate.Gate(gate_name="test",
parent="test parent",
x="X",
y="Y",
method="manual",
sampling=kwargs)
data = pd.DataFrame({
"X": np.random.normal(1, scale=0.5, size=1000),
"Y": np.random.normal(1, scale=0.5, size=1000)
})
sample = g._downsample(data=data)
if kwargs.get("method") is None:
assert sample is None
else:
assert sample.shape[0] < data.shape[0]
def test_transform_none():
g = gate.Gate(gate_name="test",
parent="test parent",
x="X",
y="Y",
method="manual")
data = pd.DataFrame({
"X": np.random.normal(1, scale=0.5, size=1000),
"Y": np.random.normal(1, scale=0.5, size=1000)
})
transformed = g._transform(data)
assert isinstance(transformed, pd.DataFrame)
assert transformed.shape[0] == 1000
assert transformed.shape[1] == 2
for i in ["X", "Y"]:
assert transformed[i].mean() == pytest.approx(1., 0.1)
assert transformed[i].std() == pytest.approx(0.5, 0.1)
def test_transform_x():
g = gate.Gate(gate_name="test",
parent="test parent",
x="X",
y="Y",
method="manual",
transformations={"x": "logicle"})
data = pd.DataFrame({
"X": np.random.normal(1, scale=0.5, size=1000),
"Y": np.random.normal(1, scale=0.5, size=1000)
})
transformed = g._transform(data)
assert isinstance(transformed, pd.DataFrame)
assert transformed.shape[0] == 1000
assert transformed.shape[1] == 2
assert transformed["X"].mean() != pytest.approx(1., 0.1)
assert transformed["X"].std() != pytest.approx(0.5, 0.1)
assert transformed["Y"].mean() == pytest.approx(1., 0.1)
assert transformed["Y"].std() == pytest.approx(0.5, 0.1)
def test_dim_reduction():
g = gate.Gate(gate_name="test",
parent="test parent",
x="X",
y="Y",
method="manual",
dim_reduction={
"method": "UMAP",
"n_neighbors": 100
})
data = pd.DataFrame({
"X": np.random.normal(1, 0.5, 1000),
"Y": np.random.normal(1, 0.5, 1000),
"Z": np.random.normal(1, 0.5, 1000),
"W": np.random.normal(1, 0.5, 1000)
})
data = g._dim_reduction(data=data)
assert g.x == "UMAP1"
assert g.y == "UMAP2"
assert data.shape == (1000, 6)
assert all([f"UMAP{i + 1}" in data.columns for i in range(2)])