Python* API Reference for Intel® Data Analytics Acceleration Library 2019

sgd_mini_dense_batch.py

1 # file: sgd_mini_dense_batch.py
2 #===============================================================================
3 # Copyright 2014-2018 Intel Corporation.
4 #
5 # This software and the related documents are Intel copyrighted materials, and
6 # your use of them is governed by the express license under which they were
7 # provided to you (License). Unless the License provides otherwise, you may not
8 # use, modify, copy, publish, distribute, disclose or transmit this software or
9 # the related documents without Intel's prior written permission.
10 #
11 # This software and the related documents are provided as is, with no express
12 # or implied warranties, other than those that are expressly stated in the
13 # License.
14 #===============================================================================
15 
16 #
17 # ! Content:
18 # ! Python example of the Stochastic gradient descent algorithm
19 # !*****************************************************************************
20 
21 #
22 
23 
24 #
25 
26 import os
27 import sys
28 
29 import numpy as np
30 
31 import daal.algorithms.optimization_solver as optimization_solver
32 import daal.algorithms.optimization_solver.mse
33 import daal.algorithms.optimization_solver.sgd
34 import daal.algorithms.optimization_solver.iterative_solver
35 
36 from daal.data_management import (
37  DataSourceIface, FileDataSource, HomogenNumericTable, MergedNumericTable, NumericTableIface
38 )
39 
40 utils_folder = os.path.realpath(os.path.abspath(os.path.dirname(os.path.dirname(__file__))))
41 if utils_folder not in sys.path:
42  sys.path.insert(0, utils_folder)
43 from utils import printNumericTable
44 
45 datasetFileName = os.path.join('..', 'data', 'batch', 'mse.csv')
46 
47 nFeatures = 3
48 accuracyThreshold = 0.0000001
49 nIterations = 1000
50 batchSize = 4
51 learningRate = 0.5
52 initialPoint = np.array([[8], [2], [1], [4]], dtype=np.float64)
53 
54 if __name__ == "__main__":
55 
56  # Initialize FileDataSource<CSVFeatureManager> to retrieve the input data from a .csv file
57  dataSource = FileDataSource(datasetFileName,
58  DataSourceIface.notAllocateNumericTable,
59  DataSourceIface.doDictionaryFromContext)
60 
61  # Create Numeric Tables for data and values for dependent variable
62  data = HomogenNumericTable(nFeatures, 0, NumericTableIface.doNotAllocate)
63  dependentVariables = HomogenNumericTable(1, 0, NumericTableIface.doNotAllocate)
64  mergedData = MergedNumericTable(data, dependentVariables)
65 
66  # Retrieve the data from the input file
67  dataSource.loadDataBlock(mergedData)
68 
69  nVectors = data.getNumberOfRows()
70 
71  mseObjectiveFunction = optimization_solver.mse.Batch(nVectors)
72  mseObjectiveFunction.input.set(optimization_solver.mse.data, data)
73  mseObjectiveFunction.input.set(optimization_solver.mse.dependentVariables, dependentVariables)
74 
75  # Create objects to compute the Stochastic gradient descent result using the mini-batch method
76  sgdMiniBatchAlgorithm = optimization_solver.sgd.Batch(mseObjectiveFunction, method=optimization_solver.sgd.miniBatch)
77 
78  # Set input objects for the the Stochastic gradient descent algorithm
79  sgdMiniBatchAlgorithm.input.setInput(optimization_solver.iterative_solver.inputArgument,
80  HomogenNumericTable(initialPoint))
81  sgdMiniBatchAlgorithm.parameter.learningRateSequence = HomogenNumericTable(1, 1, NumericTableIface.doAllocate,
82  learningRate)
83  sgdMiniBatchAlgorithm.parameter.nIterations = nIterations
84  sgdMiniBatchAlgorithm.parameter.batchSize = batchSize
85  sgdMiniBatchAlgorithm.parameter.accuracyThreshold = accuracyThreshold
86 
87  # Compute the Stochastic gradient descent result
88  # Result class from daal.algorithms.optimization_solver.iterative_solver
89  res = sgdMiniBatchAlgorithm.compute()
90 
91  # Print computed the Stochastic gradient descent result
92  printNumericTable(res.getResult(optimization_solver.iterative_solver.minimum), "Minimum")
93  printNumericTable(res.getResult(optimization_solver.iterative_solver.nIterations), "Number of iterations performed:")

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