Aprori and FP growth shows wrong itemsets

[jahidhasanlinix]

I was following the mlxtend docs to do the experiment on Apriori and FP growth and compare it but it seems I am getting different itemsets where it should be the same as I use the same df. Any suggestions why this occurring?

[rasbt]

Thanks for sharing your concern. One thing we could look at is if the arrays maybe just have a different ordering. Could you try comparing the sorted DataFrames? E.g., by

apriori_itemsets.sort_values('itemsets')[:5]

and

fp_itemsets.sort_values('itemsets')[:5]

[jahidhasanlinix]

Thank you for your response. So I tried with the sort_values() It did change the itemsets in the output but both itemsets are not the same, it's still different. Is there any suggestion afterward to fix this issue?

[rasbt]

Hm sorry, could you sort by support?

[jahidhasanlinix]

Still the same. I am not sure why this, it should be the same as I followed the documentation correctly... Its like they have same support, but itemsets are different in both cases.

[jahidhasanlinix]

I tried like this, this gives some matched itemsets but still not accurate as it gives different index numbers even some matches in both cases.
**apriori_itemsets.sort_values(['support','itemsets'], ascending=[True, False])[:5]**

[rasbt]

yeah it should give the same itemsets. I am not sure what's going on either.

[jahidhasanlinix]

So finally I able to resolved it. .sort_values(by=['support','itemsets'], ascending=False) made this changes helped me to get same itemsets. Thank you for your suggestions, it really helped me to solve this.

[rasbt]

Oh glad to hear! Sorry to hear about the frustration behind it, but it's nice to know that you solved it (and that there's no bug :)).

That's actually super useful, too! I am thinking whether the sorting should maybe be part of the apriori & fpgrowth functions (or at least mentioned in the docs). I will add it to the issue tracker.

[jahidhasanlinix]

Thank you for your response. Yeah, I guess it would be nice if the sorting is a part of of these algorithms.

[rasbt]

Sounds good! I will try to add it soon once the semester calms down a bit :)