def __make_one_dataset(self, vasp_dir):
test = VASP_DataExtract(vasp_dir=vasp_dir)
test.get_atom_and_position_info()
a = test.get_output_as_atom3Dspace()
if len(a.atoms_pos_info) <= 4: # 如果样本不够，这是很可能出现的
print_file("No enough samples for %s, which have %s." %
(vasp_dir, len(a.atoms_pos_info)))
del self.total_info[vasp_dir]
return
print_file("vasp_dir %s have sample %s" %
(vasp_dir, len(a.atoms_pos_info)))
self.total_info[vasp_dir]["generated"] = 1
# 这里的x y不是坐标而是坐标x和能量y
self.total_info[vasp_dir]['x'], self.total_info[vasp_dir][
'y'], atom_cases = a.generate_data()
self.atom_cases = self.atom_cases.union(atom_cases)
print("AtomCases", self.atom_cases)
self.total_info[vasp_dir]['atom_cases'] = self.atom_cases
def from_vasp_dir_and_energy_list(vasp_dirs,
slab_energy=None,
final_ads_energy=None,
only_x=False,
description=""):
'''
if use slab energy, y will be energy of every step - slab energy
if use final_energy, y will be energy of every step - energy of final step + final energy
'''
if only_x == False:
assert slab_energy is not None or final_ads_energy is not None, "At least feed one reference energy"
if (slab_energy is not None and final_ads_energy is not None):
raise ValueError("Can only feed one type of energy!")
if slab_energy is not None: use_slab_energy = True
else: use_slab_energy = False
coordinate = []
energy = []
box_tensor = []
print("\nNow extracting data from Vasp ... ")
for i in tqdm.trange(len(vasp_dirs)):
vasp_dir = vasp_dirs[i]
vde = VASP_DataExtract(vasp_dir)
coord, _, _ = vde.get_output_as_atom3Dspace().generate_data()
e = vde.get_energy_info()
vde.get_box_tensor_and_type_info()
_box_tensor = vde.box_tensor_info
_t = []
for j in range(len(e)):
_t.append(e[j + 1])
e = _t
coordinate.append(coord)
if only_x == False:
if use_slab_energy:
energy.append(np.array(e) - float(slab_energy[i]))
else:
energy.append(
np.array(e) - float(e[-1]) +
float(final_ads_energy[i]))
box_tensor.append(_box_tensor)
if only_x == True:
return Dataset(coordinate,
energy,
box_tensor,
only_x=True,
description=description)
else:
return Dataset(coordinate,
energy,
box_tensor,
only_x=False,
description=description)
def generate_vasp_dir_energy_table(vasp_dir, to_csv=False):
'''
give a dir contains many vasp dirs, like the dir named Vasp, that contrains: Pt_OH1, Pt_OH2, Pt_OH3 ...
and generate a excel/csv table like:
Vasp Dirs | slab energy
Pt_OH1 | 0
Pt_OH2 | 0
Pt_OH3 | 0
and then you need to fill the slab energy
'''
vasp_dirs = VASP_DataExtract.get_all_VASP_dirs_in_dir(vasp_dir)
energy = [0 for _ in range(len(vasp_dirs))]
t = np.array([vasp_dirs, energy])
t = t.T
data = pd.DataFrame(t)
data.stack(level=-1)
data.columns = ["Vasp Dirs", "slab energy"]
if to_csv:
data.to_csv("vasp_dir_path_energy_table.csv", index=False)
else:
data.to_excel("vasp_dir_path_energy_table.xlsx", index=False)
from VDE.VASPMoleculeFeature import VASP_DataExtract
from soapml.SOAPTransformer import SOAPTransformer
test = VASP_DataExtract("/home/yb/Desktop/Dataset/碳纳米管掺杂/整理后/5-5/b/h/3")
a = test.get_output_as_atom3Dspace()
coord, energy, atom_cases = a.generate_data()
atom_cases = set(atom_cases)
atom_cases.update([5, 7, 8]) # add B and N
atom_cases.remove(1) # remove H
#atom_cases.remove(6) # remove C
transformer = SOAPTransformer(encode_atom_cases=atom_cases)
for i in range(coord.shape[0]):
sample = coord[-1, :, :]
x = transformer.transform(sample, [[0, 0, 0]])
print(x.shape)
print(x[..., 0])
print(...)
exit()
np.save(save_file, self.total_encoded_grid)
return self.total_encoded_grid
def debug_encoded_grid(self, file_name="encoded_grid.npy"):
temp = np.load(file_name)
print(temp[0, 0, 0])
print(temp[0, 0, temp.shape[2] - 1])
print(temp[0, temp.shape[1] - 1, temp.shape[2] - 1])
print(temp[temp.shape[0] - 1, temp.shape[1] - 1, temp.shape[2] - 1])
def load_encoded_grid(self, file_name):
return np.load(file_name)
if __name__ == '__main__':
a = VASP_DataExtract(vasp_dir="data\Pd_CH_s_fcc_stand.cif")
c = a.get_output_as_atom3Dspace()
e = a.get_energy_info()
coord, energy, atom_case = c.generate_data()
# 1 得到坐标的atom case的信息，输入Vasp dir
# print(coord)
a = AtomGrid(coord, atom_case)
# 2 得到边界信息，然后根据边界人为确定格子大小和边界
a.get_grid_border()
a.make_grid(minX=-6,
maxX=15,
minY=-6,
maxY=15,
def encode_method_test():
a = VASP_DataExtract(vasp_dir="data\Pd_CH_s_fcc_stand.cif")
c = a.get_output_as_atom3Dspace()
e = a.get_energy_info()
coord, energy, atom_case = c.generate_data()
a = AtomGrid(coord, atom_case)
a.get_grid_border()
a.make_grid(minX=-0.5,
maxX=7,
minY=-3,
maxY=7.5,
minZ=-0.5,
maxZ=9.5,
resolutionX=100,
resolutionY=100,
resolutionZ=100)
train_index = [1]
en2 = a.grid_encode(train_index)[0]
en1 = a.grid_encode1(train_index)[0]
# en2 方法比en1快很多很多，同时要比较一下编码结果是否一致
print(en1)
print("_______________________")
print(en2)
print(en1 - en2)
print(np.sum(en1 - en2))
def make_dataset(vasp_dir, atom_case, save=False, max_dataset_number=-1):
a = VASP_DataExtract(vasp_dir=vasp_dir)
vasp_dir_name = vasp_dir.split("/")[-1].split('\\')[-1]
c = a.get_output_as_atom3Dspace()
e = a.get_energy_info()
coord, energy, _atom_case = c.generate_data()
for i in _atom_case:
if i not in atom_case:
raise ValueError(
"Input Atom Case Didn't Contain %s, make sure all atom cases are in atom_case"
% _atom_case)
atom_case = atom_case
# 1 得到坐标的atom case的信息，输入Vasp dir
# print(coord)
a = AtomGrid(coord, atom_case)
# 2 得到边界信息，然后根据边界人为确定格子大小和边界
a.get_grid_border()
a.make_grid(minX=-6,
maxX=15,
minY=-6,
maxY=15,
minZ=-6,
maxZ=15,
resolutionX=60,
resolutionY=60,
resolutionZ=60)
# 3 进行编码，给出编码的index，打乱分数据集,然后存储不同的数据集
all_index = list(range(coord.shape[0]))[:max_dataset_number]
np.random.shuffle(all_index)
print(all_index)
sample_num = len(all_index)
train_test_ratio = 0.7
trainX = testX = None
train_index = all_index[:int(sample_num * train_test_ratio)]
test_index = all_index[int(sample_num * train_test_ratio):]
if (save != False):
a.grid_encode(train_index, "%s_trainX.npy" % vasp_dir_name)
a.grid_encode(test_index, "%s_testX.npy" % vasp_dir_name)
else:
trainX = a.grid_encode(train_index, save=False)
testX = a.grid_encode(test_index, save=False)
trainY = []
testY = []
for i in train_index:
trainY.append(energy[i])
for i in test_index:
testY.append(energy[i])
trainY = np.array(trainY)
testY = np.array(testY)
if save != False:
np.save("%s_trainY.npy" % vasp_dir_name, trainY)
np.save("%s_testY.npy" % vasp_dir_name, testY)
else:
return trainX, trainY, testX, testY
if __name__ == '__main__':
# TODO: 增加交替训练，这很重要！！
all_true_y = []
all_pred_y = []
atom_case = (1, 6, 8, 78)
np.random.seed(1)
a = GridConv3D(60, 60, 60, 4) # 四种元素，先训练CHOPt都有的
a.build()
all_vasp_dir = VASP_DataExtract.get_all_VASP_dirs_in_dir(
os.getcwd()) # 获得当前的VASP文件夹作为训练集
index = 0
for vasp_dir in all_vasp_dir:
index += 1
print("train for %s" % vasp_dir)
print("Total Process: %s/%s" % (index, len(all_vasp_dir)))
# 选择不保存
trainX, trainY, testX, testY = make_dataset(
vasp_dir, save=False, max_dataset_number=50,
atom_case=atom_case) # 最多50个sample拿来训练
# 如果保存可以载入
#trainX,trainY,testX,testY = load_dataset(vasp_dir)
a.fit(trainX, trainY, epochs=1, batch_size=1)
for vasp_dir in all_vasp_dir: