def test_prechecking(self):
target = "K_VRH"
af = AutoFeaturizer(preset="best")
df = self.test_df[['composition', target]]
af.fit(df, target)
classes = [f.__class__.__name__ for f in af.featurizers["composition"]]
self.assertNotIn("YangSolidSolution", classes)
self.assertNotIn("Miedema", classes)
self.assertIn("ElementProperty", classes)
def test_use_metaselector(self):
# Test to see if metaselector works for this dataset
df = copy.copy(self.test_df.iloc[:self.limit])
target = "K_VRH"
af = AutoFeaturizer()
af.fit(df, target)
self.assertIsNotNone(af.metaselector)
dataset_mfs = af.metaselector.dataset_mfs
self.assertIn("composition_metafeatures", dataset_mfs.keys())
self.assertIn("structure_metafeatures", dataset_mfs.keys())
self.assertIsNotNone(dataset_mfs["composition_metafeatures"])
self.assertIsNotNone(dataset_mfs["structure_metafeatures"])
comp_mfs = dataset_mfs["composition_metafeatures"]
self.assertEqual(comp_mfs["number_of_compositions"], 5)
self.assertAlmostEqual(comp_mfs["percent_of_all_metal"], 0.2)
self.assertAlmostEqual(comp_mfs["percent_of_metal_nonmetal"], 0.8)
self.assertAlmostEqual(comp_mfs["percent_of_all_nonmetal"], 0.0)
self.assertAlmostEqual(comp_mfs["percent_of_contain_trans_metal"], 0.8)
self.assertEqual(comp_mfs["number_of_different_elements"], 7)
self.assertAlmostEqual(comp_mfs["avg_number_of_elements"], 2.2)
self.assertEqual(comp_mfs["max_number_of_elements"], 3)
self.assertEqual(comp_mfs["min_number_of_elements"], 1)
struct_mfs = dataset_mfs["structure_metafeatures"]
self.assertEqual(struct_mfs["number_of_structures"], 5)
self.assertAlmostEqual(struct_mfs["percent_of_ordered_structures"],
1.0)
self.assertAlmostEqual(struct_mfs["avg_number_of_sites"], 7.0)
self.assertEqual(struct_mfs["max_number_of_sites"], 12)
self.assertEqual(
struct_mfs["number_of_different_elements_in_structures"], 7)
excludes = af.metaselector.excludes
self.assertIn("IonProperty", excludes)
self.assertIn("Miedema", excludes)
self.assertIn("OxidationStates", excludes)
self.assertIn("YangSolidSolution", excludes)
self.assertIn("TMetalFraction", excludes)
self.assertIn("ElectronegativityDiff", excludes)
self.assertIn("CationProperty", excludes)
self.assertIn("ElectronAffinity", excludes)
df = af.fit_transform(df, target)
ef = ElectronAffinity()
ef_feats = ef.feature_labels()
self.assertFalse(any([f in df.columns for f in ef_feats]))
self.assertFalse(any([f in df.columns for f in ef_feats]))
def test_transferability(self):
"""
Test that an autofeaturizer object is able to be fit on one dataset
and applied to another.
"""
target = "K_VRH"
cols = ["composition", target]
df1 = self.test_df[cols].iloc[:self.limit]
df2 = self.test_df[cols].iloc[-1 * self.limit:]
af = AutoFeaturizer()
af.fit(df1, target)
df2 = af.transform(df2, target)
self.assertAlmostEqual(df2[target].iloc[0], 111.788114, places=5)
self.assertAlmostEqual(df2["minimum X"].iloc[1], 1.36, places=2)
def test_prechecking(self):
target = "K_VRH"
af = AutoFeaturizer(preset="express")
df = self.test_df[["composition", target]]
# Increase the minimum precheck fraction for purposes of this test
af.min_precheck_frac = 0.99
af.fit(df, target)
classes = [f.__class__.__name__ for f in af.featurizers["composition"]]
# both of these should be around 0.922 precheck fraction, so they fail
# the precheck test.
self.assertNotIn("YangSolidSolution", classes)
self.assertNotIn("Miedema", classes)
# ElementProperty precheck is correct for all entries, so it should pass
self.assertIn("ElementProperty", classes)
def test_exclude_by_users(self):
"""
Test custom args for featurizers to use.
"""
df = copy.copy(self.test_df.iloc[:self.limit])
target = "K_VRH"
exclude = ["ElementProperty"]
ep = ElementProperty.from_preset("matminer")
ep_feats = ep.feature_labels()
# Test to make sure excluded does not show up
af = AutoFeaturizer(exclude=exclude, preset="fast")
af.fit(df, target)
df = af.fit_transform(df, target)
self.assertTrue(af.auto_featurizer)
self.assertIn("ElementProperty", af.exclude)
self.assertFalse(any([f in df.columns for f in ep_feats]))