Fix sign of xi

molpal/acquirer/metrics.py

@@ -227,7 +227,7 @@ def ei(Y_mean: np.ndarray, Y_var: np.ndarray, current_max: float, xi: float = 0.
     E_imp : np.ndarray
         the expected improvement acquisition scores
     """
-    I = Y_mean - current_max + xi
+    I = Y_mean - current_max - xi
     Y_sd = np.sqrt(Y_var)
     with np.errstate(divide="ignore", invalid="ignore"):
         Z = I / Y_sd
@@ -255,7 +255,7 @@ def pi(Y_mean: np.ndarray, Y_var: np.ndarray, current_max: float, xi: float = 0.
     P_imp : np.ndarray
         the probability of improvement acquisition scores
     """
-    I = Y_mean - current_max + xi
+    I = Y_mean - current_max - xi
     with np.errstate(divide="ignore"):
         Z = I / np.sqrt(Y_var)
     P_imp = norm.cdf(Z)

molpal/models/chemprop/data/utils.py

@@ -25,7 +25,7 @@ def preprocess_smiles_columns(
     :return: The preprocessed version of :code:`smiles_column` which is guaranteed to be a list.
     """
     smiles_columns = smiles_columns if smiles_columns is not None else [None]
-    smiles_columns = [smiles_columns] if type(smiles_columns) != list else smiles_columns
+    smiles_columns = [smiles_columns] if not isinstance(smiles_columns, list) else smiles_columns
 
     return smiles_columns
 

molpal/models/chemprop/features/featurization.py

@@ -141,7 +141,7 @@ def __init__(self, mol: Union[str, Chem.Mol], atom_descriptors: np.ndarray = Non
         :param mol: A SMILES or an RDKit molecule.
         """
         # Convert SMILES to RDKit molecule if necessary
-        if type(mol) == str:
+        if isinstance(mol, str):
             mol = Chem.MolFromSmiles(mol)
 
         self.n_atoms = 0  # number of atoms

molpal/models/chemprop/utils.py

@@ -116,7 +116,7 @@ def accuracy(
     :param threshold: The threshold above which a prediction is a 1 and below which (inclusive) a prediction is a 0.
     :return: The computed accuracy.
     """
-    if type(preds[0]) == list:  # multiclass
+    if isinstance(preds[0], list):  # multiclass
         hard_preds = [p.index(max(p)) for p in preds]
     else:
         hard_preds = [1 if p > threshold else 0 for p in preds]  # binary prediction

tests/test_metrics.py

@@ -82,14 +82,14 @@ def test_thompson(Y_mean, Y_var, stochastic):
 def test_ei(Y_mean: np.ndarray, Y_var_0: np.ndarray, xi, curr_max):
     U = metrics.ei(Y_mean, Y_var_0, curr_max, xi)
 
-    np.testing.assert_array_less(0, U[Y_mean + xi > curr_max])
-    np.testing.assert_array_less(U[Y_mean + xi <= curr_max], 0)
+    np.testing.assert_array_less(0, U[Y_mean - xi > curr_max])
+    np.testing.assert_array_less(U[Y_mean - xi <= curr_max], 0)
 
 def test_pi(Y_mean: np.ndarray, Y_var_0: np.ndarray, xi, curr_max):
     U = metrics.pi(Y_mean, Y_var_0, curr_max, xi)
 
-    np.testing.assert_array_equal(1, U[Y_mean + xi > curr_max])
-    np.testing.assert_array_equal(0, U[Y_mean + xi <= curr_max])
+    np.testing.assert_array_equal(1, U[Y_mean - xi > curr_max])
+    np.testing.assert_array_equal(0, U[Y_mean - xi <= curr_max])
 
 def test_threshold(Y_mean: np.ndarray, threshold: float):
     U = metrics.threshold(Y_mean, threshold)