def _split_timeline(self, size: float, one_ts: bool) -> None:
"""
Split an event log dataframe by time to peform split-validation.
prefered method time splitting removing incomplete traces.
If the testing set is smaller than the 10% of the log size
the second method is sort by traces start and split taking the whole
traces no matter if they are contained in the timeframe or not
Parameters
----------
size : float, validation percentage.
one_ts : bool, Support only one timestamp.
"""
# Split log data
splitter = ls.LogSplitter(self.log.data)
# train, valdn = splitter.split_log('random', size, one_ts)
train, valdn = splitter.split_log('timeline_contained', size, one_ts)
total_events = len(self.log.data)
# Check size and change time splitting method if necesary
if len(valdn) < int(total_events * 0.1):
train, valdn = splitter.split_log('timeline_trace', size, one_ts)
# Set splits
key = 'end_timestamp' if one_ts else 'start_timestamp'
valdn = pd.DataFrame(valdn)
train = pd.DataFrame(train)
# If the log is big sample train partition
train = self._sample_log(train)
# Save partitions
self.log_valdn = (valdn.sort_values(
key, ascending=True).reset_index(drop=True))
self.log_train = copy.deepcopy(self.log)
self.log_train.set_data(
train.sort_values(
key, ascending=True).reset_index(drop=True).to_dict('records'))
def _split_timeline(self, size: float, one_ts: bool) -> None:
"""
Split an event log dataframe by time to peform split-validation.
prefered method time splitting removing incomplete traces.
If the testing set is smaller than the 10% of the log size
the second method is sort by traces start and split taking the whole
traces no matter if they are contained in the timeframe or not
Parameters
----------
size : float, validation percentage.
one_ts : bool, Support only one timestamp.
"""
# Split log data
splitter = ls.LogSplitter(self.log)
train, valdn = splitter.split_log('timeline_contained', size, one_ts)
total_events = len(self.log)
# Check size and change time splitting method if necesary
if len(valdn) < int(total_events * 0.1):
train, valdn = splitter.split_log('timeline_trace', size, one_ts)
# Set splits
key = 'end_timestamp' if one_ts else 'start_timestamp'
valdn = pd.DataFrame(valdn)
train = pd.DataFrame(train)
valdn = valdn[~valdn.task.isin(['Start', 'End'])]
train = train[~train.task.isin(['Start', 'End'])]
self.log_valdn = (valdn.sort_values(
key, ascending=True).reset_index(drop=True))
self.log_train = (train.sort_values(
key, ascending=True).reset_index(drop=True))
def split_log_test():
# Event log reading
column_names = {
'Case ID': 'caseid',
'Activity': 'task',
'lifecycle:transition': 'event_type',
'Resource': 'user'
}
settings = {
'timeformat': '%Y-%m-%dT%H:%M:%S.%f',
'column_names': column_names,
'one_timestamp': False,
'filter_d_attrib': True
}
log = lr.LogReader(
os.path.join('tests', 'fixtures', 'PurchasingExample.xes'), settings)
splitter = tl.LogSplitter(log.data)
train, test = splitter.split_log('timeline_contained', 0.8,
settings['one_timestamp'])
print(len(log.data))
print(len(train))
print(len(test))
log_min = pd.DataFrame(log.data).start_timestamp.min()
log_max = pd.DataFrame(log.data).end_timestamp.max()
train_min = pd.DataFrame(train).start_timestamp.min()
train_max = pd.DataFrame(train).end_timestamp.max()
test_min = pd.DataFrame(test).start_timestamp.min()
test_max = pd.DataFrame(test).end_timestamp.max()
print(log_min)
print(train_min)
print(train_max)
print(test_min)
print(test_max)
print(log_max)
assert (log_min == train_min)
assert (log_max == test_max)
assert (train_max < log_max)
assert (train_max < test_max)
assert (train_max < test_min)
print('##################')
splitter2 = tl.LogSplitter(train)
train, test = splitter2.split_log('timeline_contained', 0.8,
settings['one_timestamp'])
print(len(train))
print(len(test))
print('##################')
train, test = splitter.split_log('timeline_trace', 0.8,
settings['one_timestamp'])
print(len(log.data))
print(len(train))
print(len(test))
log_min = pd.DataFrame(log.data).start_timestamp.min()
log_max = pd.DataFrame(log.data).end_timestamp.max()
train_min = pd.DataFrame(train).start_timestamp.min()
train_max = pd.DataFrame(train).end_timestamp.max()
test_min = pd.DataFrame(test).start_timestamp.min()
test_max = pd.DataFrame(test).end_timestamp.max()
print(log_min)
print(train_min)
print(train_max)
print(test_min)
print(test_max)
print(log_max)
assert (log_min == train_min)
assert (log_max == test_max)
assert (train_max < log_max)
assert (train_max < test_max)
print('##################')
print(len(pd.DataFrame(log.data).caseid.unique()))
splitter3 = tl.LogSplitter(log.data)
train, test = splitter3.split_log('random', 0.8, settings['one_timestamp'])
print(len(train.caseid.unique()))
print(len(test.caseid.unique()))