def remove_previous_checkpoint(model_dir: str, port: Port, start_time: str, current_time: str,
new_optimum: bool, base_port: str = None) -> None:
"""
Remove a previous checkpoint's model- and checkpoint-file. Identification via file_type and encoded start_time:
Files with the same start_time as given in the parameters are considered to be the same training
:param model_dir: Directory to models for given port (including port name)
:param port: Port
:param start_time: Main identification for files that belong to the same training within the given directory
:param current_time: Current time for identifying older checkpoints
:param new_optimum: Specify if new optimum (= model) was found, so previous model gets removed
:param base_port: Base port name if checkpoint is made during transfer
:return: None
"""
current_time = as_datetime(current_time)
cp_type = "checkpoint-base" if base_port is None else "checkpoint-transfer"
model_type = "model-base" if base_port is None else "model-transfer"
for file in os.listdir(model_dir):
_, ext = os.path.splitext(file)
if file.startswith(cp_type) or (new_optimum and file.startswith(model_type)):
if ext in [".pt", ".tar"]: # model or checkpoint file
_, cp_port_name, cp_start_time, cp_end_time, _ = \
decode_model_file(file) if ext == ".pt" else decode_checkpoint_file(file)
cp_end_time = as_datetime(cp_end_time)
if cp_port_name == port.name and cp_start_time == start_time and cp_end_time < current_time:
os.remove(os.path.join(model_dir, file))
def __init__(self, parent, defn):
super(File, self).__init__(parent, 'files/{file_id}', defn)
if 'person_info' in defn:
person_info, to, frm = process_person_info(parent,
defn['person_info'],
defn['addresses'])
self.person_info = person_info
self.addresses = {'to': to, 'from': frm}
self.date = as_datetime(self.date)
def __init__(self, parent, defn):
super(File, self).__init__(parent, 'files/{file_id}', defn)
if 'person_info' in defn:
person_info, to, frm = process_person_info(parent, defn['person_info'], defn['addresses'])
self.person_info = person_info
self.addresses = {
'to': to,
'from': frm
}
self.date = as_datetime(self.date)
def find_latest_dataset_config_path(dataset_dir: str, training_type: str) -> str:
if training_type not in ["base", "transfer"]:
raise ValueError(f"Unknown training type '{training_type}'. Not in [base, transfer]")
max_time, path = None, None
for file in os.listdir(dataset_dir):
if file.startswith(f"dataset-config-{training_type}"):
_, start_time = decode_dataset_config_file(file)
start_time = as_datetime(start_time)
if max_time is None or max_time < start_time:
max_time = start_time
path = file
return path if path is None else os.path.join(dataset_dir, path)
def __init__(self, parent, defn):
super(Message, self).__init__(parent, 'messages/{message_id}', defn)
person_info, to, frm = process_person_info(parent, defn['person_info'],
defn['addresses'])
self.person_info = person_info
self.addresses = {'to': to, 'from': frm}
self.date = as_datetime(self.date)
if 'files' in defn:
self.files = [File(self.parent, f) for f in defn['files']]
if 'headers' in defn:
self.process_headers(defn['headers'])
def __init__(self, parent, defn):
super(Message, self).__init__(parent, 'messages/{message_id}', defn)
person_info, to, frm = process_person_info(parent, defn['person_info'], defn['addresses'])
self.person_info = person_info
self.addresses = {
'to': to,
'from': frm
}
self.date = as_datetime(self.date)
if 'files' in defn:
self.files = [File(self.parent, f) for f in defn['files']]
if 'headers' in defn:
self.process_headers(defn['headers'])
def get_files(self, **params):
params = Resource.sanitize_params(params, ['limit', 'offset'])
body = self.request_uri('files', params=params)
result = []
for obj in body:
file_name = obj.get('file_name')
occurrences = []
for o in obj.get('occurrences'):
occurrences.append({
'file': File(self.parent, o),
'message': Message(self.parent, o)
})
result.append({
'file_name': obj.get('file_name'),
'latest_date': as_datetime(obj.get('latestDate')),
'occurrences': occurrences
})
return result
