Python* API Reference for Intel® Data Analytics Acceleration Library 2018 Update 2

kmeans_csr_distr.py

1 # file: kmeans_csr_distr.py
2 #===============================================================================
3 # Copyright 2014-2018 Intel Corporation
4 # All Rights Reserved.
5 #
6 # If this software was obtained under the Intel Simplified Software License,
7 # the following terms apply:
8 #
9 # The source code, information and material ("Material") contained herein is
10 # owned by Intel Corporation or its suppliers or licensors, and title to such
11 # Material remains with Intel Corporation or its suppliers or licensors. The
12 # Material contains proprietary information of Intel or its suppliers and
13 # licensors. The Material is protected by worldwide copyright laws and treaty
14 # provisions. No part of the Material may be used, copied, reproduced,
15 # modified, published, uploaded, posted, transmitted, distributed or disclosed
16 # in any way without Intel's prior express written permission. No license under
17 # any patent, copyright or other intellectual property rights in the Material
18 # is granted to or conferred upon you, either expressly, by implication,
19 # inducement, estoppel or otherwise. Any license under such intellectual
20 # property rights must be express and approved by Intel in writing.
21 #
22 # Unless otherwise agreed by Intel in writing, you may not remove or alter this
23 # notice or any other notice embedded in Materials by Intel or Intel's
24 # suppliers or licensors in any way.
25 #
26 #
27 # If this software was obtained under the Apache License, Version 2.0 (the
28 # "License"), the following terms apply:
29 #
30 # You may not use this file except in compliance with the License. You may
31 # obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0
32 #
33 #
34 # Unless required by applicable law or agreed to in writing, software
35 # distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
36 # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
37 #
38 # See the License for the specific language governing permissions and
39 # limitations under the License.
40 #===============================================================================
41 
42 
43 
44 
45 import os
46 import sys
47 
48 import daal.algorithms.kmeans as kmeans
49 import daal.algorithms.kmeans.init as init
50 from daal import step1Local, step2Master
51 
52 utils_folder = os.path.realpath(os.path.abspath(os.path.dirname(os.path.dirname(__file__))))
53 if utils_folder not in sys.path:
54  sys.path.insert(0, utils_folder)
55 from utils import printNumericTable, createSparseTable
56 
57 DAAL_PREFIX = os.path.join('..', 'data')
58 
59 # K-Means algorithm parameters
60 nClusters = 20
61 nIterations = 5
62 nBlocks = 4
63 nVectorsInBlock = 8000
64 
65 dataFileNames = [
66  os.path.join(DAAL_PREFIX, 'batch', 'kmeans_csr.csv'),
67  os.path.join(DAAL_PREFIX, 'batch', 'kmeans_csr.csv'),
68  os.path.join(DAAL_PREFIX, 'batch', 'kmeans_csr.csv'),
69  os.path.join(DAAL_PREFIX, 'batch', 'kmeans_csr.csv')
70 ]
71 
72 dataTable = [0] * nBlocks
73 
74 if __name__ == "__main__":
75 
76  masterAlgorithm = kmeans.Distributed(step2Master, nClusters, method=kmeans.lloydCSR, )
77 
78  centroids = None
79  assignments = [0] * nBlocks
80 
81  masterInitAlgorithm = init.Distributed(step2Master, nClusters, method=init.randomDense)
82 
83  for i in range(nBlocks):
84 
85  # Read dataFileNames and create a numeric table to store the input data
86  dataTable[i] = createSparseTable(dataFileNames[i])
87 
88  # Create an algorithm object for the K-Means algorithm
89  localInit = init.Distributed(step1Local, nClusters, nBlocks * nVectorsInBlock, i * nVectorsInBlock, method=init.randomDense)
90 
91  localInit.input.set(init.data, dataTable[i])
92  # compute and add input for next
93  masterInitAlgorithm.input.add(init.partialResults, localInit.compute())
94 
95  masterInitAlgorithm.compute()
96  res = masterInitAlgorithm.finalizeCompute()
97  centroids = res.get(init.centroids)
98 
99  for it in range(nIterations):
100  for i in range(nBlocks):
101  # Create an algorithm object for the K-Means algorithm
102  localAlgorithm = kmeans.Distributed(step1Local, nClusters, it == nIterations, method=kmeans.lloydCSR)
103 
104  # Set the input data to the algorithm
105  localAlgorithm.input.set(kmeans.data, dataTable[i])
106  localAlgorithm.input.set(kmeans.inputCentroids, centroids)
107 
108  pres = localAlgorithm.compute()
109 
110  masterAlgorithm.input.add(kmeans.partialResults, pres)
111 
112  masterAlgorithm.compute()
113  result = masterAlgorithm.finalizeCompute()
114 
115  centroids = result.get(kmeans.centroids)
116  objectiveFunction = result.get(kmeans.objectiveFunction)
117 
118  for i in range(nBlocks):
119  # Create an algorithm object for the K-Means algorithm
120  localAlgorithm = kmeans.Batch(nClusters, 0, method=kmeans.lloydCSR)
121 
122  # Set the input data to the algorithm
123  localAlgorithm.input.set(kmeans.data, dataTable[i])
124  localAlgorithm.input.set(kmeans.inputCentroids, centroids)
125 
126  res = localAlgorithm.compute()
127 
128  assignments[i] = res.get(kmeans.assignments)
129 
130  # Print the clusterization results
131  printNumericTable(assignments[0], "First 10 cluster assignments from 1st node:", 10)
132  printNumericTable(centroids, "First 10 dimensions of centroids:", 20, 10)
133  printNumericTable(objectiveFunction, "Objective function value:")

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