def __fit_gmcm_using_matlab(self,
gmcm_fitting_k: int,
gmcm_max_fitting_iteration: int,
gmcm_fitting_attempt: int,
debug: bool = False):
"""
用MATLAB拟合GMCM，并储存数据
:param gmcm_fitting_k: GMCM的component数量
:param gmcm_max_fitting_iteration: GMCM的max fitting iteration
:return:
"""
if gmcm_fitting_attempt == 1:
# eng = matlab.engine.start_matlab()
eng.addpath(self.matlab_script_folder_path, nargout=0)
if not try_to_find_file(self.gmcm_model_file_):
eng.estimate_gmcm_and_save(double(
self.ndarray_data_in_uniform.tolist()),
self.gmcm_model_file_,
gmcm_fitting_k,
gmcm_max_fitting_iteration,
nargout=0)
if debug:
self.plot_ndarray_data_in_uniform_and_simulated(1_000_000)
# eng.quit()
else:
for i in range(gmcm_fitting_attempt):
if i > 0:
renamed = self.gmcm_model_file_.split(".mat")
save_name = renamed[0] + "_" + str(i) + ".mat"
else:
save_name = self.gmcm_model_file_
# eng = matlab.engine.start_matlab()
eng.addpath(self.matlab_script_folder_path, nargout=0)
if not try_to_find_file(save_name):
eng.estimate_gmcm_and_save(double(
self.ndarray_data_in_uniform.tolist()),
save_name,
gmcm_fitting_k,
gmcm_max_fitting_iteration,
nargout=0)
protect = self.gmcm_model_file_
if i > 0:
save_name = self.gmcm_model_file_.split(
f".mat")[0] + "_" + str(i) + ".mat"
self.gmcm_model_file_ = save_name
if debug:
if i == 0:
self.plot_ndarray_data_in_uniform_and_simulated(
200_000, title=f"attempt = {i} ")
else:
self.plot_simulated(200_000, title=f"attempt = {i} ")
self.gmcm_model_file_ = protect
def energies_paper_train_nilm_models_for_ampds2_dataset(top_n: int = 3):
"""
只从ampds2_dataset中的training_set中选取top_n个（默认3个）appliance。分别是HPE, FRE and CDE。
这里是符合nilmtk标准的算法。
torch的算法单独处理
:return:
"""
# 准备训练数据
training_set, test_set, _ = get_training_set_and_test_set_for_ampds2_dataset(
)
top_n_train_elec = training_set.select_using_appliances(
original_name=['HPE', 'FRE', 'CDE'])
# 模型save的路径
models_path = Path('../../../Data/Results/Energies_paper/Ampds2')
models_path.mkdir(parents=True, exist_ok=True)
# 训练所有模型
models = {'CO': CombinatorialOptimisation(), 'FHMM': FHMM()}
sample_period = 60 # 都是down sample到60s
for this_model_name, this_model in models.items():
this_model_file = models_path / (this_model_name + '.pkl')
if try_to_find_file(this_model_file):
this_model.import_model(this_model_file)
else:
print("*" * 20)
print(this_model_name)
print("*" * 20)
start = time.time()
this_model.train(top_n_train_elec, sample_period=sample_period)
this_model.export_model(this_model_file)
end = time.time()
print("Runtime =", end - start, "seconds.")
def load_npy_file(file_path: Path):
"""
载入np文件。有的话就返回，如果没有的话则返回None
"""
file_path = str(file_path)
if try_to_find_file(file_path) is False:
return None
else:
return np.load(file_path)
def load_pkl_file(file_path: Path):
if not re.match(r".*\.pkl$", str(file_path)):
raise Exception("'file_path' should end with '.pkl'")
file_path = str(file_path)
if try_to_find_file(file_path) is False:
return None
else:
with open(file_path, "rb") as f:
return pickle.load(f)
def fit(self,
only_at_edge_idx: int = None,
gmcm_fitting_attempt: int = 1,
*,
gmcm_fitting_k: int = 8):
# fit模型
initialised_pair_copula_of_each_edge = []
for edge_idx, this_edge_gmcm in enumerate(
self.gmcm_model_files_for_construction):
if only_at_edge_idx is not None:
if edge_idx > only_at_edge_idx:
continue
# 没有的话就fit并且写入ndarray_data_like_in_uniform信息
if not try_to_find_file(this_edge_gmcm):
if self.resolved_construction['conditioning'][edge_idx] == ():
# edge_var_1 和 edge_var_2分别代表两个考虑的变量。至此，还是1代表第一维度，等下生成gmcm的时候的索引要减去1
edge_var_1, edge_var_2 = self.resolved_construction[
'conditioned'][edge_idx]
input_left = self.ndarray_data_in_uniform[:,
edge_var_1 - 1]
input_right = self.ndarray_data_in_uniform[:,
edge_var_2 - 1]
else:
(input_left_copula_idx, input_right_copula_idx,
left_conditioning, right_conditioning
) = self.identify_input_copula_for_tree(edge_idx)
# 计算左右输入
input_left = initialised_pair_copula_of_each_edge[
input_left_copula_idx].cal_copula_cdf_partial_derivative(
use_ndarray_data_in_uniform=True,
partial_derivative_var_idx=(left_conditioning, ))
input_right = initialised_pair_copula_of_each_edge[
input_right_copula_idx].cal_copula_cdf_partial_derivative(
use_ndarray_data_in_uniform=True,
partial_derivative_var_idx=(right_conditioning, ))
initialised_pair_copula_of_each_edge.append(
GMCM(gmcm_model_file_=this_edge_gmcm,
ndarray_data_in_uniform=self.all_vars_valid_data(
np.stack((input_left, input_right), axis=1)),
gmcm_fitting_k=gmcm_fitting_k,
gmcm_max_fitting_iteration=50_000,
gmcm_fitting_attempt=gmcm_fitting_attempt,
debug=True,
str_name='GMCM_{}'.format(
str(self.resolved_construction['conditioned']
[edge_idx]) + '|' +
str(self.resolved_construction['conditioning']
[edge_idx]))))
# 重新修正ndarray_data_in_uniform使其包含nan，size对齐输入
initialised_pair_copula_of_each_edge[
-1].ndarray_data_in_uniform = np.stack(
(input_left, input_right), axis=1)
def pair_copula_instance_of_each_edge(self):
gmcm_models = []
for edge_idx, this_edge_gmcm in enumerate(
self.gmcm_model_files_for_construction):
if try_to_find_file(this_edge_gmcm):
gmcm_models.append(
GMCM(gmcm_model_file_=this_edge_gmcm,
str_name='GMCM_{}'.format(
str(self.resolved_construction['conditioned']
[edge_idx]) + '|' +
str(self.resolved_construction['conditioning']
[edge_idx]))))
else:
gmcm_models.append(None)
return tuple(gmcm_models)
def update_exist_pkl_file_otherwise_run_and_save(file_path: Path):
if not re.match(r".*\.pkl$", str(file_path)):
raise Exception("'file_path' should end with '.pkl'")
assert try_to_find_file(file_path)
def decorator(func):
@functools.wraps(func)
def wrapper(*args, **kwargs):
existing_file = load_pkl_file(file_path)
obj = func(*args, existing_file=existing_file, **kwargs)
save_pkl_file(file_path, obj)
return obj
return wrapper
return decorator