1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
|
# All Rights Reserved
# This code is published under the GNU Lesser General Public License (GNU LGPL)
import StOptSDDP
import StOptGeners
import Simulators
import SDDPOptimizers
import imp
## @package backwardForwardSDDP
## backward and forward sweep by SDDP
## Permits to execute forward and backward sweep in python
## This is the python version of backwardForwardSDDP.cpp
## author Xavier Warin
## @brief Achieve forward and backward sweep by SDDP
## @param p_optimizer defines the optimiser necessary to optimize a step for one simulation solving a LP
## @param p_nbSimulCheckForSimu defines the number of simulations to check convergence
## @param p_initialState initial state at the beginning of simulation
## @param p_finalCut object of final cuts
## @param p_dates vector of exercised dates, last dates correspond to the final cut object
## @param p_meshForReg number of mesh for regression in each direction
## @param p_nameRegressor name of the archive to store regressors
## @param p_nameCut name of the archive to store cuts
## @param p_nameVisitedStates name of the archive to store visited states
## @param p_iter maximum iteration of SDDP, on return the number of iterations achieved
## @param p_accuracy accuracy asked , on return estimation of accuracy achieved (expressed in %)
## @param p_nStepConv every p_nStepConv convergence is checked
## @return backward and forward valorization
def backwardForwardSDDP(p_optimizer, p_nbSimulCheckForSimu, p_initialState, p_finalCut,p_dates,
p_meshForReg, p_nameRegressor, p_nameCut, p_nameVisitedStates,
p_iter, p_accuracy,p_nStepConv):
try:
imp.find_module('mpi4py')
Found = True
except:
print("Not parallel module found ")
Found = False
if Found:
from mpi4py import MPI
world = MPI.COMM_WORLD
simulatorForOptim = p_optimizer.getSimulatorBackward()
simulatorForSim = p_optimizer.getSimulatorForward()
if Found:
bTask = (world.rank==0)
else:
bTask = True
if bTask:
# archive for regressors
archiveForRegressor =StOptGeners.BinaryFileArchive(p_nameRegressor,"w")
# archive of first set of admissible states
archiveForInitialState = StOptGeners.BinaryFileArchive(p_nameVisitedStates,"w")
# list of states
vecSetOfStates= [None] * (p_dates.size-1)
# create regressors for each date except the last
for idate in range(p_dates.size-2,0,-1):
simulatorForOptim.updateDateIndex(idate)
# initial regressor
particlesCurrent = simulatorForOptim.getParticles()
# regressor
regCurrent =StOptSDDP.LocalLinearRegressionForSDDP(False,particlesCurrent,p_meshForReg)
# store the regressor
archiveForRegressor.dump(regCurrent)
setOfStates = StOptSDDP.SDDPVisitedStates(regCurrent.getNbMeshTotal())
setOfStates.addVisitedStateForAll(p_optimizer.oneAdmissibleState(p_dates[idate + 1]),regCurrent)
# ad admissible state
vecSetOfStates[idate] =setOfStates
simulatorForOptim.updateDateIndex(0)
particlesInit = simulatorForOptim.getParticles()
regInit= StOptSDDP.LocalLinearRegressionForSDDP(True,particlesInit,p_meshForReg)
archiveForRegressor.dump(regInit)
setOfStates = StOptSDDP.SDDPVisitedStates(1)
setOfStates.addVisitedStateForAll(p_initialState,regInit)
vecSetOfStates[0]= setOfStates
for idate in range(0,p_dates.size-1):
archiveForInitialState.dump(vecSetOfStates[idate])
iterMax = p_iter;
p_iter = 0;
accuracy = p_accuracy;
p_accuracy = 1e10;
# archive to read regressor : currently all processor reads the regression
archiveReadRegressor = StOptGeners.BinaryFileArchive(p_nameRegressor, "r");
# only create for first task
archiveForCuts = StOptGeners.BinaryFileArchive(p_nameCut, "w+");
# to store all backward values
backwardValues =[]
# forward value
forwardValueForConv = 0.
istep = 0;
# store evolution of convergence
backwardMinusForwardPrev = 0;
while ((accuracy < p_accuracy) and (p_iter < iterMax)):
# increase step
istep += 1;
# actualize time for simulators
simulatorForOptim.resetTime()
simulatorForSim.resetTime()
if Found:
world.Barrier()
# backward sweep
backwardValues.append(StOptSDDP.backwardSDDP(p_optimizer , simulatorForOptim, p_dates, p_initialState,
p_finalCut, archiveReadRegressor,
p_nameVisitedStates , archiveForCuts))
if Found:
world.Barrier()
if ((istep == p_nStepConv) or (p_iter == 0)):
istep = 0
oldParticleNb = simulatorForSim.getNbSimul()
simulatorForSim.updateSimulationNumberAndResetTime(p_nbSimulCheckForSimu)
bIncreaseCut = False;
forwardValueForConv = StOptSDDP.forwardSDDP(p_optimizer , simulatorForSim, p_dates, p_initialState, p_finalCut, bIncreaseCut, archiveReadRegressor,
archiveForCuts, p_nameVisitedStates)
p_accuracy = abs((backwardValues[p_iter] - forwardValueForConv) / forwardValueForConv)
simulatorForSim.updateSimulationNumberAndResetTime(oldParticleNb)
if bTask:
print(" ACCURACY ", p_accuracy, " Backward ", backwardValues[p_iter], " Forward ", forwardValueForConv , "p_iter " ,p_iter , " accuracy " , p_accuracy )
backwardMinusForward = backwardValues[p_iter] - forwardValueForConv;
if (p_iter > 0):
if (backwardMinusForward * backwardMinusForwardPrev < 0):
if bTask:
print(" Curve are crossing : increase sample and simulations to get more accurate solution, decrease step for checking convergence")
#exit
p_accuracy = 0.
backwardMinusForwardPrev = backwardMinusForward;
elif(p_iter>0):
backwardMinusForward = backwardValues[p_iter] - forwardValueForConv;
if (backwardMinusForward * backwardMinusForwardPrev < 0):
if bTask:
print(" Curve are crossing : increase sample and simulations to get more accurate solution, decrease step for checking convergence")
#exit
p_accuracy = 0.
p_iter += 1;
if Found:
world.Barrier()
return [backwardValues[p_iter - 1], forwardValueForConv]
|
