def delete_duplicate_results(result_folder):
res_pool = ResultPool()
res_pool.load_results(result_folder)
var_params = res_pool.varying_params()
unique_var_params = []
duplicate_ids = []
all_result_file_names = res_pool.result_file_names()
for i_exp, params in enumerate(var_params):
if np.any([dict_equal(params, p) for p in unique_var_params]):
log.warn("Duplicate result {:s}".format(all_result_file_names[i_exp]))
duplicate_ids.append(i_exp)
else:
unique_var_params.append(params)
# Delete result/experiment/model(outdated, used to exist)/param files
for i_exp in duplicate_ids:
result_file_name = all_result_file_names[i_exp]
yaml_file_name = result_file_name.replace('.result.pkl', '.yaml')
model_file_name = result_file_name.replace('.result.pkl', '.pkl')
model_param_file_name = result_file_name.replace('.result.pkl', '.npy')
delete_if_exists(result_file_name)
delete_if_exists(yaml_file_name)
delete_if_exists(model_file_name)
delete_if_exists(model_param_file_name)
def delete_duplicate_results(result_folder):
res_pool = ResultPool()
res_pool.load_results(result_folder)
var_params = res_pool.varying_params()
unique_var_params = []
duplicate_ids = []
all_result_file_names = res_pool.result_file_names()
for i_exp, params in enumerate(var_params):
if np.any([dict_equal(params, p) for p in unique_var_params]):
log.warn("Duplicate result {:s}".format(
all_result_file_names[i_exp]))
duplicate_ids.append(i_exp)
else:
unique_var_params.append(params)
# Delete result/experiment/model(outdated, used to exist)/param files
for i_exp in duplicate_ids:
result_file_name = all_result_file_names[i_exp]
yaml_file_name = result_file_name.replace('.result.pkl', '.yaml')
model_file_name = result_file_name.replace('.result.pkl', '.pkl')
model_param_file_name = result_file_name.replace('.result.pkl', '.npy')
delete_if_exists(result_file_name)
delete_if_exists(yaml_file_name)
delete_if_exists(model_file_name)
delete_if_exists(model_param_file_name)
def mark_duplicate_results(result_folder, tag_dict):
res_pool = ResultPool()
res_pool.load_results(result_folder)
var_params = res_pool.varying_params()
unique_var_params = []
duplicate_ids = []
all_result_file_names = res_pool.result_file_names()
for i_exp, params in enumerate(var_params):
if np.any([dict_equal(params, p) for p in unique_var_params]):
log.warn("Duplicate result {:s}".format(all_result_file_names[i_exp]))
duplicate_ids.append(i_exp)
else:
unique_var_params.append(params)
# Update parameters
for i_exp in duplicate_ids:
result_file_name = all_result_file_names[i_exp]
result = np.load(result_file_name)
result.parameters.update(tag_dict)
pickle.dump(result, open(result_file_name, 'w'))
def _skip_already_done_experiments(self):
log.info("Check if some experiments were already run...")
clean_all_train_strs = []
# Go through all experiments, all result folders
# and check if experiments already exist
# First collect all folder paths and load results
# in order not to load results twice
# Possible optimization: First get all save paths
# Then only load results once for respective save paths
all_folder_paths = []
for i_experiment in range(len(self._all_train_strs)):
folder_path = self._create_save_folder_path(i_experiment)
all_folder_paths.append(folder_path)
unique_folder_paths = set(all_folder_paths)
folder_path_to_results = dict()
for folder_path in unique_folder_paths:
existing_result_files = glob(folder_path + "*[0-9].result.pkl")
results = [np.load(f) for f in existing_result_files]
folder_path_to_results[folder_path] = results
for i_experiment in range(len(self._all_train_strs)):
train_str = self._all_train_strs[i_experiment]
train_dict = self._load_without_layers(train_str)
original_params = train_dict['original_params']
folder_path = all_folder_paths[i_experiment]
results = folder_path_to_results[folder_path]
experiment_already_ran = False
for r in results:
if dict_equal(r.parameters, original_params):
experiment_already_ran = True
log.warn("Already ran id {:d} {:s}".format(
i_experiment, str(original_params)))
if not experiment_already_ran:
clean_all_train_strs.append(train_str)
self._all_train_strs = clean_all_train_strs
def _skip_already_done_experiments(self):
log.info("Check if some experiments were already run...")
clean_all_train_strs = []
# Go through all experiments, all result folders
# and check if experiments already exist
# First collect all folder paths and load results
# in order not to load results twice
# Possible optimization: First get all save paths
# Then only load results once for respective save paths
all_folder_paths = []
for i_experiment in range(len(self._all_train_strs)):
folder_path = self._create_save_folder_path(i_experiment)
all_folder_paths.append(folder_path)
unique_folder_paths = set(all_folder_paths)
folder_path_to_results = dict()
for folder_path in unique_folder_paths:
existing_result_files = glob(folder_path + "*[0-9].result.pkl")
results = [np.load(f) for f in existing_result_files]
folder_path_to_results[folder_path] = results
for i_experiment in range(len(self._all_train_strs)):
train_str = self._all_train_strs[i_experiment]
train_dict = self._load_without_layers(train_str)
original_params = train_dict['original_params']
folder_path = all_folder_paths[i_experiment]
results = folder_path_to_results[folder_path]
experiment_already_ran = False
for r in results:
if dict_equal(r.parameters, original_params):
experiment_already_ran = True
log.warn("Already ran id {:d} {:s}".format(i_experiment,
str(original_params)))
if not experiment_already_ran:
clean_all_train_strs.append(train_str)
self._all_train_strs = clean_all_train_strs
def mark_duplicate_results(result_folder, tag_dict):
res_pool = ResultPool()
res_pool.load_results(result_folder)
var_params = res_pool.varying_params()
unique_var_params = []
duplicate_ids = []
all_result_file_names = res_pool.result_file_names()
for i_exp, params in enumerate(var_params):
if np.any([dict_equal(params, p) for p in unique_var_params]):
log.warn("Duplicate result {:s}".format(
all_result_file_names[i_exp]))
duplicate_ids.append(i_exp)
else:
unique_var_params.append(params)
# Update parameters
for i_exp in duplicate_ids:
result_file_name = all_result_file_names[i_exp]
result = np.load(result_file_name)
result.parameters.update(tag_dict)
pickle.dump(result, open(result_file_name, 'w'))
