Further improve generate_new_combinations

The apriori-gen function described in section 2.1.1 of Apriori paper
has two steps; the first step had been implemented in 96dfd4d.

The second step of apriori-gen function is called prune step, it takes
candidates c from first step and check that all (k-1) tuples built by
removing any single element from c is in L(k-1).

Efficient lookups require some dedicated data structure; Apriori paper
describes how to do it with hash-trees; it is also possible to use
prefix trees (also knows as tries).

This commit uses third-party pygtrie module, to check whether this step
provides performance improvements in our case.  It can then be decided
to either keep this import or write a stripped down implementation.

mlxtend/frequent_patterns/apriori.py

@@ -6,6 +6,7 @@
 
 import numpy as np
 import pandas as pd
+import pygtrie
 from ..frequent_patterns import fpcommon as fpc
 
 
@@ -43,15 +44,24 @@ def generate_new_combinations(old_combinations):
     """
 
     length = len(old_combinations)
+    trie = pygtrie.Trie(list(zip(old_combinations, [1]*length)))
     for i, old_combination in enumerate(old_combinations):
         *head_i, _ = old_combination
         j = i + 1
         while j < length:
             *head_j, tail_j = old_combinations[j]
             if head_i != head_j:
                 break
-            yield from old_combination
-            yield tail_j
+            # Prune old_combination+(item,) if any subset is not frequent
+            candidate = tuple(old_combination) + (tail_j,)
+            for idx in range(len(candidate)):
+                test_candidate = list(candidate)
+                del test_candidate[idx]
+                if tuple(test_candidate) not in trie:
+                    # early exit from for-loop skips else clause just below
+                    break
+            else:
+                yield from candidate
             j = j + 1