Yannick's comments

Author: briling

qstack/regression/cross_validate_results.py

@@ -79,8 +79,6 @@ def cv_results(X, y,
         np.save(f"{preffix}_{n_rep}-lc-runs.npy", lc_runs)
     if save_pred:
         np_pred = np.array(predictions_n)
-        # Can not take means !!! Test-set varies with run !
-        # pred_mean = np.concatenate([np_pred.mean(axis=0),np_pred.std(axis=0)[1].reshape((1,-1))], axis=0)
         pred_mean = np.concatenate([*np_pred.reshape((n_rep, 2, -1))], axis=0)
         np.savetxt(f"{preffix}_{n_rep}-predictions.txt", pred_mean.T)
     return lc

tests/test_kernels.py

@@ -5,19 +5,23 @@
 
 
 def test_local_kernels():
-    # np.random.seed(666)
-    # X = np.random.rand(2,4)
-    # Y = np.random.rand(2,4)
-    # K_G_good = np.zeros((len(X),len(Y)))
-    # K_L_good = np.zeros((len(X),len(Y)))
-    # for i, x in enumerate(X):
-    #    for j, y in enumerate(Y):
-    #        K_G_good[i,j] = np.dot(x-y, x-y)
-    #        K_L_good[i,j] = np.sum(abs(x-y))
-    # np.exp(-K_G_good/2, out=K_G_good)
-    # np.exp(-K_L_good/2, out=K_L_good)
-    # K_dot_good = np.dot(X, Y.T)
-    # K_cos_good = K_dot_good / np.outer(np.linalg.norm(X, axis=1), np.linalg.norm(Y, axis=1))
+    """Test data is generated using:
+    ```
+    np.random.seed(666)
+    X = np.random.rand(2,4)
+    Y = np.random.rand(2,4)
+    K_G_good = np.zeros((len(X),len(Y)))
+    K_L_good = np.zeros((len(X),len(Y)))
+    for i, x in enumerate(X):
+        for j, y in enumerate(Y):
+            K_G_good[i,j] = np.dot(x-y, x-y)
+            K_L_good[i,j] = np.sum(abs(x-y))
+    np.exp(-K_G_good/2, out=K_G_good)
+    np.exp(-K_L_good/2, out=K_L_good)
+    K_dot_good = np.dot(X, Y.T)
+    K_cos_good = K_dot_good / np.outer(np.linalg.norm(X, axis=1), np.linalg.norm(Y, axis=1))
+    ```
+    """
 
     X = np.array([[0.70043712, 0.84418664, 0.67651434, 0.72785806], [0.95145796, 0.0127032 , 0.4135877 , 0.04881279]])
     Y = np.array([[0.09992856, 0.50806631, 0.20024754, 0.74415417], [0.192892  , 0.70084475, 0.29322811, 0.77447945]])

tests/test_spahm_a.py

@@ -8,12 +8,16 @@
 PATH = os.path.dirname(os.path.realpath(__file__))
 
 
-def underlying_test(true_data_relpath, X):
+def underlying_test(true_data_relpath, X, trim=False, only_z=None):
     X_true = np.load(PATH+true_data_relpath, allow_pickle=True)
-    assert (X.shape == X_true.shape)
-    for a, a_true in zip(X, X_true, strict=True):
-        assert (a[0] == a_true[0])                        # atom type
-        assert (np.linalg.norm(a[1]-a_true[1]) < 1e-08)   # atom representations
+    if only_z is not None:
+        X_true = X_true[np.isin(X_true[:,0], only_z)]
+    assert X.shape == X_true.shape
+    for (q, v), (q_true, v_true) in zip(X, X_true, strict=True):
+        assert q == q_true
+        if trim is True:
+            v = np.trim_zeros(v)  # short-version vectors should be trimmed
+        assert np.allclose(v, v_true)
 
 
 def test_water():
@@ -26,8 +30,6 @@ def test_water():
 
 def test_water_alternate():
     mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'minao', charge=0, spin=None)
-    # X = atom.get_repr(mol, ["H", "O"], None, dm=None,
-    #                  guess='LB', model='lowdin-long-x', auxbasis='ccpvdzjkfit')
     X = atom.get_repr("atom", [mol], [PATH], 'LB', spin=[None], auxbasis='ccpvdzjkfit', with_symbols=True)
     underlying_test('/data/SPAHM_a_H2O/X_H2O.npy', X)
 
@@ -37,8 +39,7 @@ def test_water_lowdinshortx():
     X = atom.get_repr("atom", [mol], [PATH+'/data/H2O.xyz'], 'LB',
                       elements=["H", "O"], spin=None, with_symbols=True,
                       model='lowdin-short-x', auxbasis='ccpvdzjkfit')
-    X = np.array([(z,np.trim_zeros(v)) for z,v in X], dtype=object)  # trimming is necessary to get the short-version vector !
-    underlying_test('/data/SPAHM_a_H2O/X_H2O_lowdin-short-x.npy', X)
+    underlying_test('/data/SPAHM_a_H2O/X_H2O_lowdin-short-x.npy', X, trim=True)
 
 
 def test_water_lowdinlong():
@@ -54,75 +55,58 @@ def test_water_lowdinshort():
     X = atom.get_repr("atom", [mol], [PATH+'/data/H2O.xyz'], 'LB',
                       elements=["H", "O"], spin=None, with_symbols=True,
                       model='lowdin-short', auxbasis='ccpvdzjkfit')
-    X = np.array([(z,np.trim_zeros(v)) for z,v in X], dtype=object)  # trimming is necessary to get the short-version vector !
-    underlying_test('/data/SPAHM_a_H2O/X_H2O_lowdin-short.npy', X)
+    underlying_test('/data/SPAHM_a_H2O/X_H2O_lowdin-short.npy', X, trim=True)
 
 
 def test_water_mr21():
     mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'minao', charge=0, spin=None)
     X = atom.get_repr("atom", [mol], [PATH+'/data/H2O.xyz'], 'LB',
                       elements=["H", "O"], spin=None, with_symbols=True,
                       model='MR2021', auxbasis='ccpvdzjkfit')
-    X = np.array([(z,np.trim_zeros(v)) for z,v in X], dtype=object)  # trimming is necessary to get the short-version vector !
-    underlying_test('/data/SPAHM_a_H2O/X_H2O_MR2021.npy', X)
+    underlying_test('/data/SPAHM_a_H2O/X_H2O_MR2021.npy', X, trim=True)
 
 
 def test_water_SAD_guess_open_shell():
-    mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'sto3g', charge=1, spin=1)  # test breaks when effective open-shell caluclation is needed
+    mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'sto3g', charge=1, spin=1)
     Xsad = atom.get_repr("atom", [mol], [PATH+'/data/H2O.xyz'], 'sad',
                          elements=["H", "O"], spin=[1], with_symbols=True,
                          xc='hf', model='sad-diff', auxbasis='ccpvdzjkfit')
     underlying_test('/data/SPAHM_a_H2O/X_H2O-RC_SAD.npy', Xsad)
 
 
 def test_water_SAD_guess_close_shell():
-    mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'sto3g', charge=0, spin=0)  # test breaks when effective open-shell caluclation is needed
+    mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'sto3g', charge=0, spin=0)
     Xsad = atom.get_repr("atom", [mol], [PATH+'/data/H2O.xyz'], 'sad',
                          elements=["H", "O"], spin=None, with_symbols=True,
                          xc='hf', model='sad-diff', auxbasis='ccpvdzjkfit')
     underlying_test('/data/SPAHM_a_H2O/X_H2O_SAD.npy', Xsad)
 
 
 def test_water_single_element():
+    only_z = ['O']
     mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'minao', charge=0, spin=None)
     X = atom.get_repr("atom", [mol], [PATH+'/data/H2O.xyz'], 'LB',
                       elements=["H", "O"], spin=None, with_symbols=True,
-                      model='lowdin-long-x', auxbasis='ccpvdzjkfit', only_z=['O'])  # requesting reps for O-atom only
-    X_true = np.load(PATH+'/data/SPAHM_a_H2O/X_H2O.npy', allow_pickle=True)
-    # the next two lines deviate from the common template
-    a = X[0]
-    assert (X.shape == np.array(X_true[0], ndmin=2).shape)
-    for a_true in X_true:
-        if a[0] == a_true[0]:                       # atom type
-            assert (np.linalg.norm(a[1]-a_true[1]) < 1e-08)   # atom representations
+                      model='lowdin-long-x', auxbasis='ccpvdzjkfit', only_z=only_z)
+    underlying_test('/data/SPAHM_a_H2O/X_H2O.npy', X, only_z=only_z)
 
 
 def test_water_single_element_short():
+    only_z = ['O']
     mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'minao', charge=0, spin=None)
     X = atom.get_repr("atom", [mol], [PATH+'/data/H2O.xyz'], 'LB',
                       elements=["H", "O"], spin=None, with_symbols=True,
-                      model='lowdin-short', auxbasis='ccpvdzjkfit', only_z=['O'])
-    X = np.array([(z,np.trim_zeros(v)) for z,v in X], dtype=object)  # trimming is necessary to get the short-version vector !
-    X_true = np.load(PATH+'/data/SPAHM_a_H2O/X_H2O_lowdin-short.npy', allow_pickle=True)
-    a = X[0]
-    assert (X.shape == np.array(X_true[0], ndmin=2).shape)
-    for a_true in X_true:
-        if a[0] == a_true[0]:                       # atom type
-            assert (np.linalg.norm(a[1]-a_true[1]) < 1e-08)   # atom representations
+                      model='lowdin-short', auxbasis='ccpvdzjkfit', only_z=only_z)
+    underlying_test('/data/SPAHM_a_H2O/X_H2O_lowdin-short.npy', X, only_z=only_z)
 
 
 def test_water_single_element_SAD():
+    only_z = ['O']
     mol = compound.xyz_to_mol(PATH+'/data/H2O.xyz', 'sto3g', charge=0, spin=0)
     X = atom.get_repr("atom", [mol], [PATH+'/data/H2O.xyz'], 'sad',
                          elements=["H", "O"], spin=None, with_symbols=True,
-                         xc='hf', model='sad-diff', auxbasis='ccpvdzjkfit', only_z=['O'])
-    X = np.array([(z,np.trim_zeros(v)) for z,v in X], dtype=object)  # trimming is necessary to get the short-version vector !
-    X_true = np.load(PATH+'/data/SPAHM_a_H2O/X_H2O_SAD.npy', allow_pickle=True)
-    a = X[0]
-    assert (X.shape == np.array(X_true[0], ndmin=2).shape)
-    for a_true in X_true:
-        if a[0] == a_true[0]:                       # atom type
-            assert (np.linalg.norm(a[1]-a_true[1]) < 1e-08)   # atom representations
+                         xc='hf', model='sad-diff', auxbasis='ccpvdzjkfit', only_z=only_z)
+    underlying_test('/data/SPAHM_a_H2O/X_H2O_SAD.npy', X, only_z=only_z)
 
 
 if __name__ == '__main__':