Github user MechCoder commented on a diff in the pull request:
https://github.com/apache/spark/pull/7854#discussion_r37826567
--- Diff: python/pyspark/mllib/linalg/__init__.py ---
@@ -461,32 +461,41 @@ def __init__(self, size, *args):
self.size = int(size)
""" Size of the vector. """
assert 1 <= len(args) <= 2, "must pass either 2 or 3 arguments"
- if len(args) == 1:
- pairs = args[0]
- if type(pairs) == dict:
- pairs = pairs.items()
- pairs = sorted(pairs)
- self.indices = np.array([p[0] for p in pairs], dtype=np.int32)
- """ A list of indices corresponding to active entries. """
- self.values = np.array([p[1] for p in pairs], dtype=np.float64)
- """ A list of values corresponding to active entries. """
+ if isinstance(args[0], bytes):
+ assert isinstance(args[1], bytes), "values should be string
too"
+ if args[0]:
+ self.indices = np.frombuffer(args[0], np.int32)
+ self.values = np.frombuffer(args[1], np.float64)
+ else:
+ # np.frombuffer() doesn't work well with empty string in
older version
+ self.indices = np.array([], dtype=np.int32)
+ self.values = np.array([], dtype=np.float64)
else:
- if isinstance(args[0], bytes):
- assert isinstance(args[1], bytes), "values should be
string too"
- if args[0]:
- self.indices = np.frombuffer(args[0], np.int32)
- self.values = np.frombuffer(args[1], np.float64)
- else:
- # np.frombuffer() doesn't work well with empty string
in older version
- self.indices = np.array([], dtype=np.int32)
- self.values = np.array([], dtype=np.float64)
+ if len(args) == 1:
+ args = args[0]
+ if isinstance(args, dict):
+ args = args.items()
+ args = list(zip(*args))
+
+ # Handle empty args case.
+ if len(args) == 0:
+ indices = []
+ values = []
else:
- self.indices = np.array(args[0], dtype=np.int32)
- self.values = np.array(args[1], dtype=np.float64)
- assert len(self.indices) == len(self.values), "index and value
arrays not same length"
- for i in xrange(len(self.indices) - 1):
- if self.indices[i] >= self.indices[i + 1]:
- raise TypeError("indices array must be sorted")
--- End diff --
Ah, I often just look at the direct SparseVector initialization but not
`Vectors.sparse`. hence I did not notice it.
The SparseVector initialization should take constant time and one should
expect the user should supply sorted indices (I remember another PR which was
closed because it did a O(n) check)
However, I do think that this should be documented somewhere clearly that
the indices provided should be sorted.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
