Merge pull request #50 from mir-group/mgp_test

changed mff name #29 & tested update_L_alpha #23

.gitignore

@@ -2,18 +2,18 @@ __pycache__/
 pwscf.save/
 .cache/
 .pytest_cache/
-/flare/tests/otf_run.out
-/flare/tests/test_outputs/*
 /tests/test_mff/Al_md.out
 /tests/test_mff/lmp_tmp/
 /test/test_mff/
-/test/mff/
 /datasets/
 /tests/test_lammps/*.npy
 /tests/test_lammps/*.txt
 /tests/ase_otf/*.alloy
 /tests/ase_otf/kv3/
 /tests/ase_otf/grid3*
+/tests/ase_otf/AgI_fast/out/
+/tests/ase_otf/
+/flare/ase/
 /tests/test_outputs/*
 /tests/*.xyz
 /docs/source/_build/html

flare/gp.py

@@ -331,41 +331,3 @@ def update_L_alpha(self):
         self.alpha = alpha
         self.ky_mat_inv = ky_mat_inv
         self.l_mat_inv = l_mat_inv
-
-    def update_L_alpha_v1(self):
-        """This function is used right after "update_db". It can update the l_mat_inv, k_mat_inv and alpha without computing the whole kernel matrix. The condition is the hyps should be frozen. See notes for derivation and calculation details."""
-        n = self.l_mat_inv.shape[0]
-        m = len(self.training_data) - n // 3  # number of data added
-        sigma_n = self.hyps[-1]
-
-        # update l_mat_inv => k_mat_inv
-        k_v = np.array([[] for i in range(n)])
-        V_mat = np.zeros((3 * m, 3 * m))
-
-        # calculate kernels for all added data
-        for i in range(m):
-            x_t = self.training_data[-1 - i]
-            k_vi = np.array([self.get_kernel_vector(x_t, d + 1)
-                             for d in range(3)]).T  # (n+3m) x 3
-            k_vi = k_vi[:n, :]
-            k_v = np.hstack([k_v, k_vi])  # n x 3m
-            for d1 in range(3):
-                for j in range(i, m):
-                    y_t = self.training_data[-1 - j]
-                    for d2 in range(3):
-                        V_mat[3 * i + d1, 3 * j + d2] = \
-                            self.kernel(x_t, y_t, d1 + 1, d2 + 1,
-                                        self.hyps, self.cutoffs)
-                        V_mat[3 * j + d2, 3 * i + d1] = V_mat[
-                            3 * i + d1, 3 * j + d2]
-        v = self.l_mat_inv @ k_v  # n x 3m
-        V_mat += sigma_n ** 2 * np.eye(3 * m) - v.T @ v
-        r_n1 = np.linalg.inv(np.linalg.cholesky(V_mat))  # 3m x 3m
-        r = - self.l_mat_inv.T @ v @ r_n1.T  # n x 3m
-        new_line1 = np.hstack([self.l_mat_inv, np.zeros((n, 3 * m))])
-        new_line2 = np.hstack([r.T, r_n1])
-        self.l_mat_inv = np.vstack([new_line1, new_line2])
-        self.ky_mat_inv = self.l_mat_inv.T @ self.l_mat_inv
-
-        # update alpha
-        self.alpha = self.ky_mat_inv @ self.training_labels_np