def test_maui_model_validates_feature_names_on_predict_after_fit():
maui_model = Maui(n_hidden=[10], n_latent=2, epochs=1)
maui_model.fit({"d1": df1, "d2": df2})
z = maui_model.transform({"d1": df1, "d2": df2})
df1_wrong_features = df1.reindex(df1.index[:len(df1.index) - 1])
with pytest.raises(ValueError):
z = maui_model.transform({"df1": df1_wrong_features, "df2": df2})
def test_maui_produces_different_prediction_when_run_twice_with_sampling():
"""This is to show the maui encoder model picks the mean of
the distribution, not a sample."""
maui_model = Maui(n_hidden=[10], n_latent=2, epochs=1)
maui_model = maui_model.fit({'d1': df1, 'd2': df2})
z1 = maui_model.transform({'d1': df1, 'd2': df2}, encoder='sample')
z2 = maui_model.transform({'d1': df1, 'd2': df2}, encoder='sample')
assert not np.allclose(z1, z2)
def test_maui_produces_same_prediction_when_run_twice():
"""This is to show the maui encoder model picks the mean of
the distribution, not a sample."""
maui_model = Maui(n_hidden=[10], n_latent=2, epochs=1)
maui_model = maui_model.fit({"d1": df1, "d2": df2})
z1 = maui_model.transform({"d1": df1, "d2": df2})
z2 = maui_model.transform({"d1": df1, "d2": df2})
assert np.allclose(z1, z2)
def test_maui_produces_pos_and_neg_zs_if_relu_embedding_false():
maui_model = Maui(n_hidden=[10],
n_latent=2,
epochs=1,
relu_embedding=False)
maui_model = maui_model.fit({"d1": df1, "d2": df2})
z1 = maui_model.transform({"d1": df1, "d2": df2})
assert not np.all(z1 >= 0)
def test_maui_updates_neural_weight_product_when_training():
maui_model = Maui(n_hidden=[10], n_latent=2, epochs=1)
z_before = maui_model.fit_transform({"d1": df1, "d2": df2})
nwp_before_fine_tuning = maui_model.get_neural_weight_product()
maui_model.fine_tune({"d1": df1, "d2": df2})
z_after = maui_model.transform({"d1": df1, "d2": df2})
nwp_after_fine_tuning = maui_model.get_neural_weight_product()
assert not np.allclose(z_before, z_after)
assert not np.allclose(nwp_before_fine_tuning, nwp_after_fine_tuning)
def test_maui_can_load_from_folder():
maui_model = Maui(n_hidden=[10], n_latent=2, epochs=1)
maui_model = maui_model.fit({"d1": df1, "d2": df2})
with tempfile.TemporaryDirectory() as tmpdirname:
maui_model.save(tmpdirname)
maui_model_from_disk = Maui.load(tmpdirname)
assert maui_model_from_disk.n_latent == maui_model.n_latent
assert np.allclose(maui_model.vae.get_weights()[0],
maui_model_from_disk.vae.get_weights()[0])
assert np.allclose(
maui_model.transform({"d1": df1, "d2": df2}),
maui_model_from_disk.transform({"d1": df1, "d2": df2})
)
def test_maui_supports_single_layer_vae():
maui_model = Maui(n_hidden=None, n_latent=2, epochs=1)
maui_model = maui_model.fit({'d1': df1, 'd2': df2})
z1 = maui_model.transform({'d1': df1, 'd2': df2})
def test_maui_produces_nonnegative_zs_if_relu_embedding_true():
maui_model = Maui(n_hidden=[10], n_latent=2, epochs=1, relu_embedding=True)
maui_model = maui_model.fit({'d1': df1, 'd2': df2})
z1 = maui_model.transform({'d1': df1, 'd2': df2})
assert np.all(z1>=0)
