Show More
@@ -1,170 +1,167 | |||
|
1 | 1 | # encoding: utf-8 |
|
2 | 2 | |
|
3 | 3 | """Classes used in scattering and gathering sequences. |
|
4 | 4 | |
|
5 | 5 | Scattering consists of partitioning a sequence and sending the various |
|
6 | 6 | pieces to individual nodes in a cluster. |
|
7 | 7 | |
|
8 | 8 | |
|
9 | 9 | Authors: |
|
10 | 10 | |
|
11 | 11 | * Brian Granger |
|
12 | 12 | * MinRK |
|
13 | 13 | |
|
14 | 14 | """ |
|
15 | 15 | |
|
16 | 16 | #------------------------------------------------------------------------------- |
|
17 | 17 | # Copyright (C) 2008-2011 The IPython Development Team |
|
18 | 18 | # |
|
19 | 19 | # Distributed under the terms of the BSD License. The full license is in |
|
20 | 20 | # the file COPYING, distributed as part of this software. |
|
21 | 21 | #------------------------------------------------------------------------------- |
|
22 | 22 | |
|
23 | 23 | #------------------------------------------------------------------------------- |
|
24 | 24 | # Imports |
|
25 | 25 | #------------------------------------------------------------------------------- |
|
26 | 26 | |
|
27 | 27 | from __future__ import division |
|
28 | 28 | |
|
29 | 29 | import types |
|
30 | 30 | from itertools import islice |
|
31 | 31 | |
|
32 | 32 | from IPython.utils.data import flatten as utils_flatten |
|
33 | 33 | |
|
34 | 34 | #------------------------------------------------------------------------------- |
|
35 | 35 | # Figure out which array packages are present and their array types |
|
36 | 36 | #------------------------------------------------------------------------------- |
|
37 | 37 | |
|
38 | 38 | arrayModules = [] |
|
39 | 39 | try: |
|
40 | 40 | import Numeric |
|
41 | 41 | except ImportError: |
|
42 | 42 | pass |
|
43 | 43 | else: |
|
44 | 44 | arrayModules.append({'module':Numeric, 'type':Numeric.arraytype}) |
|
45 | 45 | try: |
|
46 | 46 | import numpy |
|
47 | 47 | except ImportError: |
|
48 | 48 | pass |
|
49 | 49 | else: |
|
50 | 50 | arrayModules.append({'module':numpy, 'type':numpy.ndarray}) |
|
51 | 51 | try: |
|
52 | 52 | import numarray |
|
53 | 53 | except ImportError: |
|
54 | 54 | pass |
|
55 | 55 | else: |
|
56 | 56 | arrayModules.append({'module':numarray, |
|
57 | 57 | 'type':numarray.numarraycore.NumArray}) |
|
58 | 58 | |
|
59 | 59 | class Map(object): |
|
60 | 60 | """A class for partitioning a sequence using a map.""" |
|
61 | 61 | |
|
62 | def getPartition(self, seq, p, q): | |
|
63 |
"""Returns the pth partition of q partitions of seq. |
|
|
62 | def getPartition(self, seq, p, q, n=None): | |
|
63 | """Returns the pth partition of q partitions of seq. | |
|
64 | 64 |
|
|
65 | The length can be specified as `n`, | |
|
66 | otherwise it is the value of `len(seq)` | |
|
67 | """ | |
|
68 | n = len(seq) if n is None else n | |
|
65 | 69 | # Test for error conditions here |
|
66 | 70 | if p<0 or p>=q: |
|
67 | print "No partition exists." | |
|
68 | return | |
|
71 | raise ValueError("must have 0 <= p <= q, but have p=%s,q=%s" % (p, q)) | |
|
69 | 72 | |
|
70 | N = len(seq) | |
|
71 |
|
|
|
72 | basesize = N // q | |
|
73 | remainder = n % q | |
|
74 | basesize = n // q | |
|
73 | 75 | |
|
74 | 76 | if p < remainder: |
|
75 | 77 | low = p * (basesize + 1) |
|
76 | 78 | high = low + basesize + 1 |
|
77 | 79 | else: |
|
78 | 80 | low = p * basesize + remainder |
|
79 | 81 | high = low + basesize |
|
80 | 82 | |
|
81 | 83 | try: |
|
82 | 84 | result = seq[low:high] |
|
83 | 85 | except TypeError: |
|
84 | 86 | # some objects (iterators) can't be sliced, |
|
85 | 87 | # use islice: |
|
86 | 88 | result = list(islice(seq, low, high)) |
|
87 | 89 | |
|
88 | 90 | return result |
|
89 | 91 | |
|
90 | 92 | def joinPartitions(self, listOfPartitions): |
|
91 | 93 | return self.concatenate(listOfPartitions) |
|
92 | 94 | |
|
93 | 95 | def concatenate(self, listOfPartitions): |
|
94 | 96 | testObject = listOfPartitions[0] |
|
95 | 97 | # First see if we have a known array type |
|
96 | 98 | for m in arrayModules: |
|
97 | 99 | #print m |
|
98 | 100 | if isinstance(testObject, m['type']): |
|
99 | 101 | return m['module'].concatenate(listOfPartitions) |
|
100 | 102 | # Next try for Python sequence types |
|
101 | 103 | if isinstance(testObject, (types.ListType, types.TupleType)): |
|
102 | 104 | return utils_flatten(listOfPartitions) |
|
103 | 105 | # If we have scalars, just return listOfPartitions |
|
104 | 106 | return listOfPartitions |
|
105 | 107 | |
|
106 | 108 | class RoundRobinMap(Map): |
|
107 | """Partitions a sequence in a roun robin fashion. | |
|
109 | """Partitions a sequence in a round robin fashion. | |
|
108 | 110 | |
|
109 | 111 | This currently does not work! |
|
110 | 112 | """ |
|
111 | 113 | |
|
112 | def getPartition(self, seq, p, q): | |
|
113 | # if not isinstance(seq,(list,tuple)): | |
|
114 | # raise NotImplementedError("cannot RR partition type %s"%type(seq)) | |
|
115 | return seq[p:len(seq):q] | |
|
116 | #result = [] | |
|
117 | #for i in range(p,len(seq),q): | |
|
118 | # result.append(seq[i]) | |
|
119 | #return result | |
|
114 | def getPartition(self, seq, p, q, n=None): | |
|
115 | n = len(seq) if n is None else n | |
|
116 | return seq[p:n:q] | |
|
120 | 117 | |
|
121 | 118 | def joinPartitions(self, listOfPartitions): |
|
122 | 119 | testObject = listOfPartitions[0] |
|
123 | 120 | # First see if we have a known array type |
|
124 | 121 | for m in arrayModules: |
|
125 | 122 | #print m |
|
126 | 123 | if isinstance(testObject, m['type']): |
|
127 | 124 | return self.flatten_array(m['type'], listOfPartitions) |
|
128 | 125 | if isinstance(testObject, (types.ListType, types.TupleType)): |
|
129 | 126 | return self.flatten_list(listOfPartitions) |
|
130 | 127 | return listOfPartitions |
|
131 | 128 | |
|
132 | 129 | def flatten_array(self, klass, listOfPartitions): |
|
133 | 130 | test = listOfPartitions[0] |
|
134 | 131 | shape = list(test.shape) |
|
135 | 132 | shape[0] = sum([ p.shape[0] for p in listOfPartitions]) |
|
136 | 133 | A = klass(shape) |
|
137 | 134 | N = shape[0] |
|
138 | 135 | q = len(listOfPartitions) |
|
139 | 136 | for p,part in enumerate(listOfPartitions): |
|
140 | 137 | A[p:N:q] = part |
|
141 | 138 | return A |
|
142 | 139 | |
|
143 | 140 | def flatten_list(self, listOfPartitions): |
|
144 | 141 | flat = [] |
|
145 | 142 | for i in range(len(listOfPartitions[0])): |
|
146 | 143 | flat.extend([ part[i] for part in listOfPartitions if len(part) > i ]) |
|
147 | 144 | return flat |
|
148 | 145 | #lengths = [len(x) for x in listOfPartitions] |
|
149 | 146 | #maxPartitionLength = len(listOfPartitions[0]) |
|
150 | 147 | #numberOfPartitions = len(listOfPartitions) |
|
151 | 148 | #concat = self.concatenate(listOfPartitions) |
|
152 | 149 | #totalLength = len(concat) |
|
153 | 150 | #result = [] |
|
154 | 151 | #for i in range(maxPartitionLength): |
|
155 | 152 | # result.append(concat[i:totalLength:maxPartitionLength]) |
|
156 | 153 | # return self.concatenate(listOfPartitions) |
|
157 | 154 | |
|
158 | 155 | def mappable(obj): |
|
159 | 156 | """return whether an object is mappable or not.""" |
|
160 | 157 | if isinstance(obj, (tuple,list)): |
|
161 | 158 | return True |
|
162 | 159 | for m in arrayModules: |
|
163 | 160 | if isinstance(obj,m['type']): |
|
164 | 161 | return True |
|
165 | 162 | return False |
|
166 | 163 | |
|
167 | 164 | dists = {'b':Map,'r':RoundRobinMap} |
|
168 | 165 | |
|
169 | 166 | |
|
170 | 167 |
General Comments 0
You need to be logged in to leave comments.
Login now