The Data Mining Forum 
IMPORTANT: This is the old Data Mining forum.
I keep it online so that you can read the old messages.
Please post your new messages in the new forum: https://forum2.philippe-fournier-viger.com/index.php
how to decide the param of FPgrowth or Apriori
Posted by:
validf4
Date: July 18, 2014 09:25PM
use a FPgrowth or Apriori,if i set the param min support and min confidence very low, i can get many rules,if i set them high,i get fewer.
so i decide to make the programe adaptive,i run the programe repeatly,when the count of the rules is between 25 and 30,i considers the param is fit.
There is a question,i can test min support from 0.2 to 0.6,
but when the min support is 0.3 ,what is the min confidence?and when min support is 0.4 what is the min confidence?so on..
i want to find the best param,but not by guess.
so i decide to make the programe adaptive,i run the programe repeatly,when the count of the rules is between 25 and 30,i considers the param is fit.
There is a question,i can test min support from 0.2 to 0.6,
but when the min support is 0.3 ,what is the min confidence?and when min support is 0.4 what is the min confidence?so on..
i want to find the best param,but not by guess.
Re: how to decide the param of FPgrowth or Apriori
Posted by:
webmasterphilfv
Date: July 19, 2014 03:41AM
The best value for minsup is different for each dataset. For some datasets, you may get 1 million rules for minsup = 0.9. For other datasets you may need to set minsup to 0.01 to get 100 rules. There is no easy way to determine what is the best minsup value than by trying several values. But in general, the lower you set minsup, the more rules you will find and the algorithm will become slower.
For minconf, it is easier to set this parameter because it represents the confidence that you want in the rule. For example, you may not be interested to find a rule X --> Y with a confidence lower than 0.5 (50%) perhaps because it indicates that there is 50 % probability that Y appears when X appears. Since the confidence has a more clear meaning, it is usually easier to set. In some applications, I usually use values around minconf = 0.7. But you may still tune it.
Besides, another option is to use a top-k association rule mining algorithm like TopKRules. In that case, the algorithm will automatically tune minsup so that you find k rules. So for example, you can say that you want the top 1000 rules that are the most frequent with a confidence of 0.7 and the algorithm will adjust minsup automatically to find exactly these 1000 rules.
Edited 1 time(s). Last edit at 07/19/2014 03:42AM by webmasterphilfv.
For minconf, it is easier to set this parameter because it represents the confidence that you want in the rule. For example, you may not be interested to find a rule X --> Y with a confidence lower than 0.5 (50%) perhaps because it indicates that there is 50 % probability that Y appears when X appears. Since the confidence has a more clear meaning, it is usually easier to set. In some applications, I usually use values around minconf = 0.7. But you may still tune it.
Besides, another option is to use a top-k association rule mining algorithm like TopKRules. In that case, the algorithm will automatically tune minsup so that you find k rules. So for example, you can say that you want the top 1000 rules that are the most frequent with a confidence of 0.7 and the algorithm will adjust minsup automatically to find exactly these 1000 rules.
Edited 1 time(s). Last edit at 07/19/2014 03:42AM by webmasterphilfv.
Re: how to decide the param of FPgrowth or Apriori
Posted by:
validf4
Date: July 20, 2014 05:21AM
Thanks a lot,very helpful
Re: how to decide the param of FPgrowth or Apriori
Posted by:
Pooja jardosh
Date: December 18, 2014 03:43AM
I understood the problem of rule generation,that is mentioned here."if i set the param min support and min confidence very low, i can get many rules,if i set them high,i get fewer."
And you mentioned to make it adaptive.How can it be done??
and what does it mean actually??
i am not getting it.
Can anyone help?
And you mentioned to make it adaptive.How can it be done??
and what does it mean actually??
i am not getting it.
Can anyone help?