Possible instability of FastICA with all default settings
Legacy User
MemberPosts:0
Newbie
Hi,
The FastICA seems to produce different results on repeated runs with the same input data. Here is a simple example:
输入:
X1 X2 X3
1 1 1
2 3 2
3 1 3
2 3 4
1 1 5
实验:
Repeated runs of this experiment output completely different results in the log.
Are there any tricks regarding the ICA settings that would make the results reproducible?
Victor
The FastICA seems to produce different results on repeated runs with the same input data. Here is a simple example:
输入:
X1 X2 X3
1 1 1
2 3 2
3 1 3
2 3 4
1 1 5
实验:
Repeated runs of this experiment output completely different results in the log.
Are there any tricks regarding the ICA settings that would make the results reproducible?
Victor
0
Answers
thanks for pointing this out. The FastICA operator uses random number for matrix initializations, hence the different results for repeated runs. We just added a parameter "local_random_seed" which allows the control of this behaviour like we have done for all other random-based operators (like cross validation etc.). You can access the new version via CVS and of course it will also be available in the next release. All users of the RapidMiner Enterprise Edition will of course get this improvement with the next update.
Thanks again and cheers,
Ingo