saver = tf.train.Saver(tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES))
sess.run(tf.global_variables_initializer())
saver.restore(sess, "./log/model.ckpt")
for i in range(len(R_surf)):
# for Rs in R_surf[:10]:
Rl = R_surf[i] - R_Cu
dl = np.sqrt(np.sum(Rl**2, axis=-1))
dl[dl > Router] = 0
coord = np.zeros((np.sum(dl > 0), 3))
coord = Rl[dl > 0]
coord = np.concatenate((np.zeros((1, 3)), coord), axis=0)
feat = pyu.getFeat(len(coord), coord, featParams["n2b"],
featParams["n3b"])
feedDict = {tf_feat: pyf.scaleFeat(featParams, feat)}
Ei[i] = sess.run(L3, feed_dict=feedDict)
print("{}/{}".format(i, len(R_surf)), Ei[i])
# pyu.saveXYZ([coord], ["Cu"], "test.xyz")
Ei = Ei - 0.5
idx = np.zeros(len(Ei))
for k in range(3):
idx[np.abs(R_surf[:, k]) < 2] = 1
dC1O1 = np.mean(dC1O1)
dC0H = np.mean(dC0H)
aOCH = np.mean(aOCH)*np.pi/180
randomIdx = np.random.permutation(len(R_surf))[:nCase]
Rs = R_surf[randomIdx]
for iCase in range(nCase):
print(iCase)
d_sq = np.sum((Rs[iCase] - R_Cu)**2,axis=1)
Rcluster = R_Cu[d_sq < Rcut**2]
RNN = R_Cu[(d_sq < Rcut2**2) & (d_sq != 0)]
Rcluster2 = np.concatenate([(Rs[iCase])[np.newaxis,:],R_Cu[(d_sq < Rcut3**2) & (d_sq !=0)]],axis=0)
feat=pyu.getFeat(len(Rcluster2), Rcluster2-Rs[iCase],featParams['n2b'],featParams['n3b'])
E0 = pyf.getE(feat[np.newaxis,:], featParams, nnParams)[0]
print(E0)
if E0 < -0.5:
R_surfNN = pyu.getSurfNN(RNN, Rcluster, Rnb=Rcut2, chunkSize=100)
R_surf = pyu.getSurfVector(R_surfNN, Rcluster, Rnb=Rcut4, angleCutoff=angleCutoff)
feat = np.zeros([len(R_surf), featParams['nFeat']])
for iNN in range(len(R_surf)):
# print(R_surf[iNN] - Rs[iCase])
diNN_sq = np.sum((R_surf[iNN] - Rcluster)**2,axis=1)
RiNN = np.zeros((np.sum(diNN_sq < Rcut3**2), 3))
RiNN[1:] = Rcluster[(diNN_sq != 0.0 ) & (diNN_sq < Rcut3**2)]-R_surf[iNN]
# pyu.saveXYZ([RiNN], ['Cu'], 'CuOCCOH_poscar/Cu'+str(iNN)+'.xyz')
feat[iNN] = pyu.getFeat(len(RiNN), RiNN,featParams['n2b'],featParams['n3b'])