Understanding apyori's output

Question

I'm well familiar with the apriori algorithm, and the meaning of support/confidence/lift.

I'm currently using the apyori apriori implementation, and I'm not sure I understand the output of an apyori.apriori() call.

It comes out like this

> RelationRecord(items=frozenset({'item1', 'item2'}),
> support=0.15365410803449842,
> ordered_statistics=[OrderedStatistic(items_base=frozenset({'item1'}),
> items_add=frozenset({'item2'}), confidence=0.6203420891875382,
> lift=2.2233410344037092),
> OrderedStatistic(items_base=frozenset({'item2'}),
> items_add=frozenset({'item1'}), confidence=0.5507049891540131,
> lift=2.2233410344037097)])

What is the rule? There are multiple support/confidence/lift, what each one denotes?

I'd appreciate a dictionary style explanation of each part of the output

Surrounding

Answer 1

The RelationRecord reflects a subset of items, while ordered_statistics is a list of OrderedStatistics, which reflect the rules. Each OrderedStatistics' items_base is the antecedent and the items_add is the consequent. The support is stored in the RelationRecord, since it's the same for the contained rules.

In your example:

item1 -> item2 with 0.62 confidence and 2.2233410344037092x lift

item2 -> item1 with 0.55 confidence and 2.2233410344037097x lift

Both have support=0.15365410803449842.

For what it's worth, I ended up switching to using PyFIM for relative feature richness and other bundled algorithms (e.g. fp-growth).

Slipstream

Just wanted to follow up and see if my answer was helpful/you continued using apyori. If you found this helpful, please select it as the correct answer. Thanks! – Slipstream

do you know if the rules returned from the algorithm is by default sorted in some order? rules = apriori(transactions, min_support = 0.003, min_confidence = 0.2, min_lift = 3, min_length = 2) – Ralaigh

Thanks for this. I really wish apyori had a proper documentation. – Congius

@Ralaigh note that there's no keyword arg min_length. – Congius

@timegeb - Just whipped these unofficial docs together. Hope they help! zaxrosenberg.com/unofficial-apyori-documentation – Slipstream

By the way, how do I interpret a rule of the type: frozenset({}) -> fronzenset({item}) What does an empty frozenset mean? – Ho