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

lbfgs_cr_entr_loss_dense_batch.py

1 # file: lbfgs_cr_entr_loss_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 # ! Content:
17 # ! Python example of the limited memory Broyden-Fletcher-Goldfarb-Shanno
18 # ! algorithm with cross entropy loss function
19 # !*****************************************************************************
20 
21 #
22 ## <a name="DAAL-EXAMPLE-PY-LBFGS_CR_ENTR_LOSS_BATCH"></a>
23 ## \example lbfgs_cr_entr_loss_dense_batch.py
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.cross_entropy_loss
33 import daal.algorithms.optimization_solver.lbfgs
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', 'logreg_train.csv')
46 
47 nFeatures = 6
48 nClasses = 5
49 nIterations = 1000
50 stepLength = 1.0e-4
51 
52 if __name__ == "__main__":
53 
54  # Initialize FileDataSource<CSVFeatureManager> to retrieve the input data from a .csv file
55  dataSource = FileDataSource(datasetFileName,
56  DataSourceIface.notAllocateNumericTable,
57  DataSourceIface.doDictionaryFromContext)
58 
59  # Create Numeric Tables for input data and dependent variables
60  data = HomogenNumericTable(nFeatures, 0, NumericTableIface.doNotAllocate)
61  dependentVariables = HomogenNumericTable(1, 0, NumericTableIface.doNotAllocate)
62  mergedData = MergedNumericTable(data, dependentVariables)
63 
64  # Retrieve the data from input file
65  dataSource.loadDataBlock(mergedData)
66 
67  func = optimization_solver.cross_entropy_loss.Batch(nClasses, data.getNumberOfRows())
68  func.input.set(optimization_solver.cross_entropy_loss.data, data)
69  func.input.set(optimization_solver.cross_entropy_loss.dependentVariables, dependentVariables)
70 
71  # Create objects to compute LBFGS result using the default method
72  algorithm = optimization_solver.lbfgs.Batch(func)
73  algorithm.parameter.nIterations = nIterations
74  algorithm.parameter.stepLengthSequence = HomogenNumericTable(1, 1, NumericTableIface.doAllocate, stepLength)
75 
76  # Set input objects for LBFGS algorithm
77  nParameters = nClasses * (nFeatures + 1)
78  initialPoint = np.full((nParameters, 1), 0.001, dtype=np.float64)
79  algorithm.input.setInput(optimization_solver.iterative_solver.inputArgument, HomogenNumericTable(initialPoint))
80 
81  # Compute LBFGS result
82  # Result class from daal.algorithms.optimization_solver.iterative_solver
83  res = algorithm.compute()
84 
85  expectedPoint = np.array([[-2.277], [2.836], [14.985], [0.511], [7.510], [-2.831], [-5.814], [-0.033], [13.227], [-24.447], [3.730],
86  [10.394], [-10.461], [-0.766], [0.077], [1.558], [-1.133], [2.884], [-3.825], [7.699], [2.421], [-0.135], [-6.996], [1.785], [-2.294], [-9.819], [1.692],
87  [-0.725], [0.069], [-8.41], [1.458], [-3.306], [-4.719], [5.507], [-1.642]], dtype=np.float64)
88  expectedCoefficients = HomogenNumericTable(expectedPoint)
89 
90  # Print computed LBFGS results
91  printNumericTable(expectedCoefficients, "Expected coefficients:")
92  printNumericTable(res.getResult(optimization_solver.iterative_solver.minimum), "Resulting coefficients:")
93  printNumericTable(res.getResult(optimization_solver.iterative_solver.nIterations), "Number of iterations performed:")

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