Reducing H_1

[selting]

The docstring of generate_rules_apriori says:

The algorithm is based on the observation that for {a, b} -> {c, d} to
hold, both {a, b, c} -> {d} and {a, b, d} -> {c} must hold, since in
general conf( {a, b, c} -> {d} ) >= conf( {a, b} -> {c, d} ).
In other words, if either of the two one-consequent rules do not hold, then
there is no need to ever consider the two-consequent rule.

But in H_1, items are never removed if they produced a low confidence rule. This has been fixed.

NOTE: One test fails and I did not have the time to investigate exactly why:
test_generate_rules_apriori_large:
Expected :5387
Actual :5382

This must be cleared before accepting the request

[tommyod]

Since test_generate_rules_apriori_large fails, this PR changes the output of the algorithm under some inputs. Do you have a minimal example of where the current algorithm produces the wrong answer, but the PR produces the correct answer?