Bug fixes in pseudo inverse computations

.pre-commit-config.yaml

@@ -45,7 +45,11 @@ repos:
     rev: v1.5.1
     hooks:
       - id: mypy
-        additional_dependencies: [types-all]
+        additional_dependencies:
+          - types-setuptools
+          - types-paramiko
+          - types-requests
+          - types-ujson
 
 #  - repo: local
 #    hooks:

matminer/featurizers/composition/tests/test_composite.py

@@ -318,7 +318,7 @@ def test_elem_optical(self):
             self.df_nans, col_id="composition"
         )
         self.assertEqual(df_elem.isna().sum().sum(), 0)
-        self.assertAlmostEqual(df_elem.drop(columns="composition").sum().sum(), 201.3255, 4)
+        self.assertAlmostEqual(df_elem.drop(columns="composition").sum().sum(), 204.4712, 4)
 
     def test_elem_transport(self):
         df_elem = ElementProperty.from_preset("mp_transport", impute_nan=False).featurize_dataframe(
@@ -343,7 +343,7 @@ def test_elem_transport(self):
         df_elem = ElementProperty.from_preset("mp_transport", impute_nan=True).featurize_dataframe(
             self.df_nans, col_id="composition"
         )
-        self.assertAlmostEqual(df_elem.drop(columns="composition").sum().sum(), 9798095.622017656, 4)
+        self.assertAlmostEqual(df_elem.drop(columns="composition").sum().sum(), 10029874.1567, 4)
 
 
 if __name__ == "__main__":

matminer/utils/tests/test_data.py

@@ -283,7 +283,7 @@ def test_get_data(self):
         c_k = self.data_source_imputed.get_elemental_property(elem="C", property_name="k_760.0")
         self.assertAlmostEqual(c_k, 0.7462931865379264)
         og_r = self.data_source_imputed.get_elemental_property(elem="Og", property_name="R_400.0")
-        self.assertAlmostEqual(og_r, 0.4624005395190695)
+        self.assertAlmostEqual(og_r, 0.46962554794905487)
 
 
 class TestTransportData(TestCase):
@@ -308,7 +308,7 @@ def test_get_data(self):
         cu_kappan = self.data_source_imputed.get_elemental_property(elem="Cu", property_name="kappa_n")
         self.assertAlmostEqual(cu_kappan, 1814544.75663, places=5)
         og_mn = self.data_source_imputed.get_elemental_property(elem="Og", property_name="m_n")
-        self.assertAlmostEqual(og_mn, 0.03237036761677134)
+        self.assertAlmostEqual(og_mn, 0.03293018092682478)
 
 
 if __name__ == "__main__":

matminer/utils/utils.py

@@ -66,7 +66,7 @@ def get_elem_in_data(df, as_pure=False):
     else:
         for elem in Element:
             for compound in df.index.to_list():
-                if elem.name in compound and elem.name not in elems_in_df:
+                if elem in Composition(compound) and elem.name not in elems_in_df:
                     elems_in_df.append(elem.name)
 
     # Find the elements not in the data
@@ -137,5 +137,11 @@ def get_pseudo_inverse(df_init, cols=None):
     res_pi = np.vstack([res_pi, np.nan * np.ones([len(elems_not_in_df), len(df.T) - 1])])
 
     df_pi = pd.DataFrame(res_pi, columns=cols, index=pd.Index(elems_in_df + elems_not_in_df))
+    # Handle the case of hydrogen, deuterium, and tritium
+    # If all are present, there contributions are summed
+    # and given to hydrogen only. Others are removed.
+    if all(e in df_pi.index for e in ["H", "D", "T"]):
+        df_pi.loc["H"] = df_pi.loc["H"] + df_pi.loc["D"] + df_pi.loc["T"]
+        df_pi.drop(index=["T", "D"], inplace=True)
 
     return df_pi