sarray_basic_stats.py

# To perform basic math operations on each row of an SArray (or a column of an SFrame)
# it is most efficient to utilize the .apply method and specify a lambda function.
#
# For API documentation on the .apply() method, see here:
# http://turi.com/products/create/docs/generated/graphlab.SArray.apply.html#graphlab.SArray.apply

import math
import graphlab
import timeit

# create an SArray with 1,000,000 rows
sa = graphlab.SArray(range(1, 1000001))

# Log: We want to get the log for each of the rows in the SArray.
logs = sa.apply(lambda x: math.log(x))

# Getting the absolute value for each row:
pos = sa.apply(lambda x: abs(x))

# using timeit:
abs_results = timeit.timeit(stmt='[ abs(x) for x in sa ]',
                            setup='import graphlab; sa = graphlab.SArray(range(1,1000001))',
                            number=1)
glc_abs_results = timeit.timeit(stmt='sa.apply(lambda x: abs(x))',
                                setup='import graphlab; sa = graphlab.SArray(range(1,1000001))',
                                number=1)

print "Absolute value: "
print "Using [ abs(x) for x in sa ] (implicit list conversion): %g" % abs_results # laptop results: 1.6541240215301514
print "Using sa.apply(lambda x: abs(x)): %g" % glc_abs_results                    # laptop results: 0.010690927505493164

# When performing other basic stats operations, it is much more efficient to
# use the implementations provided by SArrays. This avoids implicit
# conversions to Python lists and keeps the operations in the C++ engine.
# Examples:
mean = sa.mean()
max = sa.max()
min = sa.min()
std = sa.std()

max_results = timeit.timeit(stmt='max(sa)', setup='import graphlab; sa = graphlab.SArray(range(1,1000001))', number=1)
sa_results = timeit.timeit(stmt='sa.max()', setup='import graphlab; sa = graphlab.SArray(range(1,1000001))', number=1)

print "Using max(sa): %g" % max_results         # laptop results: 1.79631s
print "Using SArray.max(): %g" % sa_results     # laptop results: 0.015002s