THPSimulation: simulated dataframe does not contains the "duration" column.

[gim4855744]

Returned dataframe of THPSimulation.simulate() does not contains duration column.
Line 678 in simulator.py, Xn.reindex() cause this problem.

Change Xn_list.append(Xn.reindex(columns=['event', 'timestamp', 'node'])) to Xn_list.append(Xn) solve the problem.

Version
1.0.4

Current output

	event	timestamp	node
0	1	6885	0
1	2	6094	0
...	...	...	...
399337	3	6924	39

Expected output

	node	event	timestamp	duration
0	0	1	7989	8.0
1	0	1	9779	12.0
...	...	...	...	...
738188	39	2	271	47.0

[shaido987]

Hi @gim4855744, thank you for raising this issue!

The current output of THPSimulation.simulate() is actually intended behavior for our use case. The TTPM algorithm only utilizes the event, timestamp, and node columns — the duration information is not needed and is therefore omitted during processing.

You can see where this is reflected in the TTPM implementation here:

trustworthyAI/gcastle/castle/algorithms/ttpm/ttpm.py

Lines 93 to 108 in bd9cac2

	def learn(self, tensor, *args, **kwargs):
	"""
	Set up and run the TTPM algorithm.
	Parameters
	----------
	tensor: pandas.DataFrame
	(V 1.0.0, we'll eliminate this constraint in the next version)
	The tensor is supposed to contain three cols:
	['event', 'timestamp', 'node']
	Description of the three columns:
	event: event name (type).
	timestamp: occurrence timestamp of event, i.e., '1615962101.0'.
	node: topological node where the event happened.
	"""

That said, if your application requires the duration column, you can modify the simulation code as you've suggested (Xn_list.append(Xn) instead of reindexing) to retain all columns.

Thanks again for your feedback!