hadoop - How to correlate all combination of arrays in an RDD? -
i have rdd model.productfeatures()
returns rdd in form of (id, array("d", (...)))
. example:
(1, array("d", (0, 1, 2))) (2, array("d", (4, 3, 2))) (3, array("d", (5, 3, 0))) ...
i calculate pairwise correlation between each array, return each id
id
array has highest correlation.
the first thing need pairs of elements, except "diagonal" they're same.
>>> rdd.cartesian(rdd).filter(lambda (x, y): x != y).collect() [((1, array('d', [0.0, 1.0, 2.0])), (2, array('d', [4.0, 3.0, 2.0]))), ((1, array('d', [0.0, 1.0, 2.0])), (3, array('d', [5.0, 3.0, 0.0]))), ((2, array('d', [4.0, 3.0, 2.0])), (1, array('d', [0.0, 1.0, 2.0]))), ((3, array('d', [5.0, 3.0, 0.0])), (1, array('d', [0.0, 1.0, 2.0]))), ((2, array('d', [4.0, 3.0, 2.0])), (3, array('d', [5.0, 3.0, 0.0]))), ((3, array('d', [5.0, 3.0, 0.0])), (2, array('d', [4.0, 3.0, 2.0])))]
then function calculate correlation , rearrange prepare last step. let's assume "correlation" mean done numpy.correlate
.
def corr_pair(((id1, a1), (id2, a2))): return id1, (id2, np.correlate(a1, a2)[0]) >>> rdd.cartesian(rdd).filter(lambda (p1, p2): p1 != p2).map(corr_pair).collect() [(1, (2, 7.0)), (1, (3, 3.0)), (2, (1, 7.0)), (3, (1, 3.0)), (2, (3, 29.0)), (3, (2, 29.0))]
to 2nd id maximum correlation each 1st id, can use reducebykey
, keep bigger one:
def keep_higher((id1, c1), (id2, c2)): if c1 > c2: return id1, c1 else: return id2, c2 >>> rdd.cartesian(rdd).filter(lambda (x, y): x != y).map(corr_pair).reducebykey(keep_higher).collect() [(1, (2, 7.0)), (2, (3, 29.0)), (3, (2, 29.0))]
Comments
Post a Comment