Python* API Reference for Intel® Data Analytics Acceleration Library 2018 Update 1

svd_dense_distr.py

1 #===============================================================================
2 # Copyright 2014-2017 Intel Corporation
3 # All Rights Reserved.
4 #
5 # If this software was obtained under the Intel Simplified Software License,
6 # the following terms apply:
7 #
8 # The source code, information and material ("Material") contained herein is
9 # owned by Intel Corporation or its suppliers or licensors, and title to such
10 # Material remains with Intel Corporation or its suppliers or licensors. The
11 # Material contains proprietary information of Intel or its suppliers and
12 # licensors. The Material is protected by worldwide copyright laws and treaty
13 # provisions. No part of the Material may be used, copied, reproduced,
14 # modified, published, uploaded, posted, transmitted, distributed or disclosed
15 # in any way without Intel's prior express written permission. No license under
16 # any patent, copyright or other intellectual property rights in the Material
17 # is granted to or conferred upon you, either expressly, by implication,
18 # inducement, estoppel or otherwise. Any license under such intellectual
19 # property rights must be express and approved by Intel in writing.
20 #
21 # Unless otherwise agreed by Intel in writing, you may not remove or alter this
22 # notice or any other notice embedded in Materials by Intel or Intel's
23 # suppliers or licensors in any way.
24 #
25 #
26 # If this software was obtained under the Apache License, Version 2.0 (the
27 # "License"), the following terms apply:
28 #
29 # You may not use this file except in compliance with the License. You may
30 # obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0
31 #
32 #
33 # Unless required by applicable law or agreed to in writing, software
34 # distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
35 # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
36 #
37 # See the License for the specific language governing permissions and
38 # limitations under the License.
39 #===============================================================================
40 
41 ## <a name="DAAL-EXAMPLE-PY-SVD_DISTRIBUTED"></a>
42 ## \example svd_dense_distr.py
43 
44 import os
45 import sys
46 
47 from daal import step1Local, step2Master, step3Local
48 from daal.algorithms import svd
49 from daal.data_management import FileDataSource, DataSourceIface
50 
51 utils_folder = os.path.realpath(os.path.abspath(os.path.dirname(os.path.dirname(__file__))))
52 if utils_folder not in sys.path:
53  sys.path.insert(0, utils_folder)
54 from utils import printNumericTable
55 
56 DAAL_PREFIX = os.path.join('..', 'data')
57 
58 # Input data set parameters
59 nBlocks = 4
60 
61 datasetFileNames = [
62  os.path.join(DAAL_PREFIX, 'distributed', 'svd_1.csv'),
63  os.path.join(DAAL_PREFIX, 'distributed', 'svd_2.csv'),
64  os.path.join(DAAL_PREFIX, 'distributed', 'svd_3.csv'),
65  os.path.join(DAAL_PREFIX, 'distributed', 'svd_4.csv')
66 ]
67 
68 dataFromStep1ForStep2 = [0] * nBlocks
69 dataFromStep1ForStep3 = [0] * nBlocks
70 dataFromStep2ForStep3 = [0] * nBlocks
71 Sigma = None
72 V = None
73 Ui = [0] * nBlocks
74 
75 
76 def computestep1Local(block):
77  global dataFromStep1ForStep2, dataFromStep1ForStep3
78 
79  # Initialize FileDataSource<CSVFeatureManager> to retrieve the input data from a .csv file
80  dataSource = FileDataSource(
81  datasetFileNames[block],
82  DataSourceIface.doAllocateNumericTable,
83  DataSourceIface.doDictionaryFromContext
84  )
85 
86  # Retrieve the input data
87  dataSource.loadDataBlock()
88 
89  # Create an algorithm to compute SVD on the local node
90  algorithm = svd.Distributed(step1Local)
91 
92  algorithm.input.set(svd.data, dataSource.getNumericTable())
93 
94  # Compute SVD and get OnlinePartialResult class from daal.algorithms.svd
95  pres = algorithm.compute()
96 
97  dataFromStep1ForStep2[block] = pres.get(svd.outputOfStep1ForStep2)
98  dataFromStep1ForStep3[block] = pres.get(svd.outputOfStep1ForStep3)
99 
100 
101 def computeOnMasterNode():
102  global Sigma, V, dataFromStep2ForStep3
103 
104  # Create an algorithm to compute SVD on the master node
105  algorithm = svd.Distributed(step2Master)
106 
107  for i in range(nBlocks):
108  algorithm.input.add(svd.inputOfStep2FromStep1, i, dataFromStep1ForStep2[i])
109 
110  # Compute SVD and get DistributedPartialResult class from daal.algorithms.svd
111  pres = algorithm.compute()
112 
113  for i in range(nBlocks):
114  dataFromStep2ForStep3[i] = pres.getCollection(svd.outputOfStep2ForStep3, i)
115 
116  res = algorithm.finalizeCompute()
117 
118  Sigma = res.get(svd.singularValues)
119  V = res.get(svd.rightSingularMatrix)
120 
121 
122 def finalizeComputestep1Local(block):
123  global Ui
124 
125  # Create an algorithm to compute SVD on the master node
126  algorithm = svd.Distributed(step3Local)
127 
128  algorithm.input.set(svd.inputOfStep3FromStep1, dataFromStep1ForStep3[block])
129  algorithm.input.set(svd.inputOfStep3FromStep2, dataFromStep2ForStep3[block])
130 
131  # Compute SVD
132  algorithm.compute()
133  res = algorithm.finalizeCompute()
134 
135  Ui[block] = res.get(svd.leftSingularMatrix)
136 
137 if __name__ == "__main__":
138 
139  for i in range(nBlocks):
140  computestep1Local(i)
141 
142  computeOnMasterNode()
143 
144  for i in range(nBlocks):
145  finalizeComputestep1Local(i)
146 
147  # Print the results
148  printNumericTable(Sigma, "Singular values:")
149  printNumericTable(V, "Right orthogonal matrix V:")
150  printNumericTable(Ui[0], "Part of left orthogonal matrix U from 1st node:", 10)

For more complete information about compiler optimizations, see our Optimization Notice.