[SPARK-4148][PySpark] fix seed distribution and add some tests for rdd.sample

The current way of seed distribution makes the random sequences from partition i and i+1 offset by 1.

~~~
In [14]: import random

In [15]: r1 = random.Random(10)

In [16]: r1.randint(0, 1)
Out[16]: 1

In [17]: r1.random()
Out[17]: 0.4288890546751146

In [18]: r1.random()
Out[18]: 0.5780913011344704

In [19]: r2 = random.Random(10)

In [20]: r2.randint(0, 1)
Out[20]: 1

In [21]: r2.randint(0, 1)
Out[21]: 0

In [22]: r2.random()
Out[22]: 0.5780913011344704
~~~

Note: The new tests are not for this bug fix.

Author: Xiangrui Meng <[email protected]>

Closes #3010 from mengxr/SPARK-4148 and squashes the following commits:

869ae4b [Xiangrui Meng] move tests tests.py
c1bacd9 [Xiangrui Meng] fix seed distribution and add some tests for rdd.sample

Author: mengxr

python/pyspark/rdd.py

@@ -316,9 +316,6 @@ def sample(self, withReplacement, fraction, seed=None):
         """
         Return a sampled subset of this RDD (relies on numpy and falls back
         on default random generator if numpy is unavailable).
-
-        >>> sc.parallelize(range(0, 100)).sample(False, 0.1, 2).collect() #doctest: +SKIP
-        [2, 3, 20, 21, 24, 41, 42, 66, 67, 89, 90, 98]
         """
         assert fraction >= 0.0, "Negative fraction value: %s" % fraction
         return self.mapPartitionsWithIndex(RDDSampler(withReplacement, fraction, seed).func, True)

python/pyspark/rddsampler.py

@@ -40,14 +40,13 @@ def __init__(self, withReplacement, seed=None):
     def initRandomGenerator(self, split):
         if self._use_numpy:
             import numpy
-            self._random = numpy.random.RandomState(self._seed)
+            self._random = numpy.random.RandomState(self._seed ^ split)
         else:
-            self._random = random.Random(self._seed)
+            self._random = random.Random(self._seed ^ split)
 
-        for _ in range(0, split):
-            # discard the next few values in the sequence to have a
-            # different seed for the different splits
-            self._random.randint(0, 2 ** 32 - 1)
+        # mixing because the initial seeds are close to each other
+        for _ in xrange(10):
+            self._random.randint(0, 1)
 
         self._split = split
         self._rand_initialized = True

python/pyspark/tests.py

@@ -648,6 +648,21 @@ def test_distinct(self):
         self.assertEquals(result.getNumPartitions(), 5)
         self.assertEquals(result.count(), 3)
 
+    def test_sample(self):
+        rdd = self.sc.parallelize(range(0, 100), 4)
+        wo = rdd.sample(False, 0.1, 2).collect()
+        wo_dup = rdd.sample(False, 0.1, 2).collect()
+        self.assertSetEqual(set(wo), set(wo_dup))
+        wr = rdd.sample(True, 0.2, 5).collect()
+        wr_dup = rdd.sample(True, 0.2, 5).collect()
+        self.assertSetEqual(set(wr), set(wr_dup))
+        wo_s10 = rdd.sample(False, 0.3, 10).collect()
+        wo_s20 = rdd.sample(False, 0.3, 20).collect()
+        self.assertNotEqual(set(wo_s10), set(wo_s20))
+        wr_s11 = rdd.sample(True, 0.4, 11).collect()
+        wr_s21 = rdd.sample(True, 0.4, 21).collect()
+        self.assertNotEqual(set(wr_s11), set(wr_s21))
+
 
 class ProfilerTests(PySparkTestCase):