def explore_dimensions(self, initialState, ranges, pointsPerDim = 10, repetitions = 3):
'''For the given vector, vary each parameter separately.
Arguments:
initialState -- where to start
ranges -- parameter ranges
pointsPerDim -- number of points along each dimension
repetitions -- how many measurements to make of each position
'''
nDimensions = len(initialState)
self.log_start()
self.log_write('RunManager.explore_dimensions, centered at %s' % prettyVec(initialState))
for dimension in range(nDimensions):
self.log_write('RunManager.explore_dimensions: dimension %d' % dimension)
points = Neighbor.uniform_spread(ranges, initialState, dimension, number = pointsPerDim, includeOrig = True)
for ii,point in enumerate(points):
for tt in range(repetitions):
print 'Iteration dimension %d, point %d, trial %d' % (dimension, ii, tt), prettyVec(point),
sys.stdout.flush()
self.log_write('%d, %d, %d, ' % (dimension, ii, tt), newline=False)
dist = self.run_robot(point)
print '%.2f' % dist
self.log_results(point, dist)
def updateResults(self, dist):
self.iterations += 1
if self.bestDist is None or dist > self.bestDist:
self.bestDist = dist
self.bestState = self.current
self.bestIter = self.iterations
print ' best (iter %3d)' % self.bestIter, prettyVec(self.bestState), '%.2f' % self.bestDist # Prints best state and distance so far
def explore_dimensions(self,
initialState,
ranges,
pointsPerDim=10,
repetitions=3):
'''For the given vector, vary each parameter separately.
Arguments:
initialState -- where to start
ranges -- parameter ranges
pointsPerDim -- number of points along each dimension
repetitions -- how many measurements to make of each position
'''
nDimensions = len(initialState)
self.log_start()
self.log_write('RunManager.explore_dimensions, centered at %s' %
prettyVec(initialState))
for dimension in range(nDimensions):
self.log_write('RunManager.explore_dimensions: dimension %d' %
dimension)
points = Neighbor.uniform_spread(ranges,
initialState,
dimension,
number=pointsPerDim,
includeOrig=True)
for ii, point in enumerate(points):
for tt in range(repetitions):
print 'Iteration dimension %d, point %d, trial %d' % (
dimension, ii, tt), prettyVec(point),
sys.stdout.flush()
self.log_write('%d, %d, %d, ' % (dimension, ii, tt),
newline=False)
dist = self.run_robot(point)
print '%.2f' % dist
self.log_results(point, dist)
def do_many_runs(self, strategy, ranges, limit=None):
if limit is None:
limit = 10000
self.log_start(strategy.logHeader())
for ii in xrange(limit):
currentState, logInfo = strategy.getNext()
print
#if hasattr(currentState, '__call__'):
# print 'Iteration %2d: %s' % (ii+1, logStr),
#else:
# print 'Iteration %2d params' % (ii+1), prettyVec(currentState),
paramStr = logInfo if isinstance(
logInfo, basestring) else prettyVec(logInfo)
print 'Iteration %2d: %s' % (ii + 1, paramStr)
sys.stdout.flush()
# Check if this state is new, and possibly skip it
#if tuple(currentState) in self.statesSoFar:
# print '*** Duplicate iteration!'
# # Skip only if using random hill climbing. In other words,
# # comment this line out if using gradient_search:
# #currentState = neighborFunction(bestState)
# continue
currentDistance = self.run_robot(currentState)
#currentDistance = input('Distance walked? ')
#self.statesSoFar.add(tuple(currentState))
#print ' walked %.2f' % currentDistance
print '%.2f' % currentDistance
strategy.updateResults(currentDistance)
#print ' best so far', prettyVec(strategy.bestState), '%.2f' % strategy.bestDist # Prints best state and distance so far
self.log_results(logInfo, currentDistance)
def do_many_runs(self, strategy, ranges, limit = None):
if limit is None:
limit = 10000
self.log_start(strategy.logHeader())
for ii in xrange(limit):
currentState, logInfo = strategy.getNext()
print
#if hasattr(currentState, '__call__'):
# print 'Iteration %2d: %s' % (ii+1, logStr),
#else:
# print 'Iteration %2d params' % (ii+1), prettyVec(currentState),
paramStr = logInfo if isinstance(logInfo, basestring) else prettyVec(logInfo)
print 'Iteration %2d: %s' % (ii+1, paramStr)
sys.stdout.flush()
# Check if this state is new, and possibly skip it
#if tuple(currentState) in self.statesSoFar:
# print '*** Duplicate iteration!'
# # Skip only if using random hill climbing. In other words,
# # comment this line out if using gradient_search:
# #currentState = neighborFunction(bestState)
# continue
currentDistance = self.run_robot(currentState)
#currentDistance = input('Distance walked? ')
#self.statesSoFar.add(tuple(currentState))
#print ' walked %.2f' % currentDistance
print '%.2f' % currentDistance
strategy.updateResults(currentDistance)
#print ' best so far', prettyVec(strategy.bestState), '%.2f' % strategy.bestDist # Prints best state and distance so far
self.log_results(logInfo, currentDistance)