rasbt / mlxtend / 2021

91%

push

travis-ci

let fpgrowth and fpmax work directly with sparse input (#622)

* let fpgrowth and fpmax work directly with sparse input

If transactions are stored in a sparse matrix, it was first converted
into a dense Numpy array in setup_fptree before processing.

This commit allows to build the FPTree directly from sparse input, which
may save memory and processing time.  This is used by both fpgrowth and
fpmax.  There is no change after FPTree is being built.

* in setup_fptree, ensure that sparse DataFrame has no zeroes

In a212806 we assumed that all values in sparse DataFrame are non null,
which should always be true.  In order to avoid this corner case, we
now call itemsets.eliminate_zeros().  The alternative would be to replace
   nonnull = itemsets.indices[itemsets.indptr[i]:itemsets.indptr[i+1]]
by
   values = itemsets.data[itemsets.indptr[i]:itemsets.indptr[i+1]]
   nonnull = itemsets.indices[itemsets.indptr[i] + values.nonzero()[0]]
but it is slower.

Add a test case.

* fix pep8 issue

1784 of 2249 branches covered (79.32%)

36 of 36 new or added lines in 3 files covered. (100.0%)

8231 of 8913 relevant lines covered (92.35%)

0.92 hits per line