Ordered itemsets

[scikit-mine]

Ordered itemsets

[lgalarra]

Thanks Rémi, that helps. I would probably recommend you to include how to run an existence test in the documentation, e.g.,
assert frozenset({14, 15, 16}) in patterns.itemset.values
It comes handy at debugging.

Cheers,
Luis

[remiadon]

Hi Luis,
Good point !!

Short answer : Yes it's possible

---

Longer answer :

I believe that having a consistent order for the output itemsets may make other tasks, such as testing

That's right, I already had some issue with inconsistent outputs with another algorithm using the same representation

For now itemsets are represented as python frozenset, mainly for 3 reasons

• a frozenset is hashable
• it's immutable
• it's a primitive data type, recognized by every Python implementation

IMO itemsets should stay hashable, it's a good practice. They also should remain immutable, to avoid opportunistic errors during the mining. The third assumption, thought, can be modified. If we import a data type from one of our dependecy, we can stick with it

My trick is to use a data type provided by pandas

>>> from pandas.core.indexes.frozen import FrozenList
>>> itemset = FrozenList([2, 3])
>>> itemset[1:]  # slicing OK
FrozenList([3])
>>> hash(itemset)  # works

Some more simple solution might be to use the tuple data type

>>> a = tuple((2, 3))
>>> a[1:]  # works
>>> hash(a) # also works

---

On the other hand, a natural ordered may not be defined for itemsets of arbitrary data types

Actually we use sortedcontainer.SortedSet to track items in LCM.fit, so we already check for items to be comparable. If not, we throw an error at .fit time, so that .transform is not even called.

Hope this helps

[remiadon]

fixed in 54a24f6

doc updated see https://scikit-mine.github.io/scikit-mine/tutorials/itemsets/LCM_on_chess.html

[lgalarra]

Thanks Rémi. I still have a recommendation. I saw the tests and now you we use the tuple notation to check the existence of an itemset. Perhaps the documentation should show the "official" way to do it (if you think that such a task may be done for other reasons besides debugging).