Пример #1
0
    def __init__(self,
                 countdown=2,
                 statistics=5,
                 improve_factor=0.5,
                 worsen_factor=1.5,
                 max_factor=1.0,
                 min_factor=1.0e-4,
                 log_size=50):

        self.improve_factor = improve_factor
        self.worsen_factor = worsen_factor
        self.max_factor = max_factor
        self.min_factor = min_factor
        self.statistics = statistics
        self.countdown_initial = countdown
        self._reset()
        self.log = CircularList(log_size)
Пример #2
0
    def __init__(self, countdown=2, statistics=5,
                 improve_factor=0.5, worsen_factor=1.5,
                 max_factor=1.0, min_factor=1.0e-4,
                 log_size=50):

        self.improve_factor = improve_factor
        self.worsen_factor = worsen_factor
        self.max_factor = max_factor
        self.min_factor = min_factor
        self.statistics = statistics
        self.countdown_initial = countdown
        self._reset()
        self.log = CircularList(log_size)
Пример #3
0
class Convergence:
    def __init__(self,
                 countdown=2,
                 statistics=5,
                 improve_factor=0.5,
                 worsen_factor=1.5,
                 max_factor=1.0,
                 min_factor=1.0e-4,
                 log_size=50):

        self.improve_factor = improve_factor
        self.worsen_factor = worsen_factor
        self.max_factor = max_factor
        self.min_factor = min_factor
        self.statistics = statistics
        self.countdown_initial = countdown
        self._reset()
        self.log = CircularList(log_size)

    def _reset(self):
        self.num = 0
        self.sum_dev_max_dmdt = 0.0
        self.sum_osc_max_dmdt = 0.0
        self.previous_max_dmdt = None
        self.countdown = self.countdown_initial

    def _quality(self, max_dmdt):
        convergence_quality = None
        if self.previous_max_dmdt != None:
            dev = max_dmdt - self.previous_max_dmdt
            osc = abs(dev)
            self.sum_osc_max_dmdt += osc
            self.sum_dev_max_dmdt += dev
            self.num += 1

            if self.num >= self.statistics and self.sum_osc_max_dmdt > 0.0:
                convergence_quality = \
                  abs(self.sum_dev_max_dmdt/self.sum_osc_max_dmdt)

                self.num = 0
                self.sum_dev_max_dmdt = 0.0
                self.sum_osc_max_dmdt = 0.0

        self.previous_max_dmdt = max_dmdt
        return convergence_quality

    def check(self, step, max_dm_dt, stopping_dm_dt, tol_factor):
        """
        Returns True when convergence has been reached.
        """
        log.debug("Entering is_converged()")
        #print "max_dm_dt=%s < %s=stopping_dm_dt ? %s" \
        #      % (max_dm_dt, stopping_dm_dt, max_dm_dt < stopping_dm_dt)
        if max_dm_dt < stopping_dm_dt:
            self.countdown -= 1
            if self.countdown < 1:
                log.debug("is_converged(): True")
                # Since we are converging, we don't want to interfere
                # and change the tolerances now! We therefore return
                # new_tol_factor = None
                self._reset()
                return True, None

            log.debug("is_converged(): Trueish (%d times converged but need "
                      "%d times to stop)" % (self.countdown, self.countdown))
            return False, None

        self.countdown = self.countdown_initial
        log.debug("is_converged(): False")

        convergence_quality = self._quality(max_dm_dt)
        self.log.append((step, max_dm_dt, stopping_dm_dt, convergence_quality))

        if convergence_quality == None:
            return False, None

        else:
            new_tol_factor = None
            if convergence_quality == 1.0:
                new_tol_factor = tol_factor * self.worsen_factor
            elif convergence_quality < 0.5:
                new_tol_factor = tol_factor * self.improve_factor
            if new_tol_factor != None:
                new_tol_factor = min(self.max_factor,
                                     max(self.min_factor, new_tol_factor))
                if new_tol_factor == tol_factor: new_tol_factor = None
        return False, new_tol_factor

    def get_log(self):
        l = self.log.get_list()
        s = "# Convergence log\n"
        s += "# step, ax dm/dt, stoppping dm/dt, conv. quality\n"
        for step, max_dm_dt, stopping_dm_dt, conv_quality in l:
            s += "%s %s %s %s\n" % (step, max_dm_dt, stopping_dm_dt,
                                    conv_quality)
        return s
Пример #4
0
class Convergence:
    def __init__(self, countdown=2, statistics=5,
                 improve_factor=0.5, worsen_factor=1.5,
                 max_factor=1.0, min_factor=1.0e-4,
                 log_size=50):

        self.improve_factor = improve_factor
        self.worsen_factor = worsen_factor
        self.max_factor = max_factor
        self.min_factor = min_factor
        self.statistics = statistics
        self.countdown_initial = countdown
        self._reset()
        self.log = CircularList(log_size)

    def _reset(self):
        self.num = 0
        self.sum_dev_max_dmdt = 0.0
        self.sum_osc_max_dmdt = 0.0
        self.previous_max_dmdt = None
        self.countdown = self.countdown_initial

    def _quality(self, max_dmdt):
        convergence_quality = None
        if self.previous_max_dmdt != None:
            dev = max_dmdt - self.previous_max_dmdt
            osc = abs(dev)
            self.sum_osc_max_dmdt += osc
            self.sum_dev_max_dmdt += dev
            self.num += 1

            if self.num >= self.statistics and self.sum_osc_max_dmdt > 0.0:
                convergence_quality = \
                  abs(self.sum_dev_max_dmdt/self.sum_osc_max_dmdt)

                self.num = 0
                self.sum_dev_max_dmdt = 0.0
                self.sum_osc_max_dmdt = 0.0

        self.previous_max_dmdt = max_dmdt
        return convergence_quality

    def check(self, step, max_dm_dt, stopping_dm_dt, tol_factor):
        """
        Returns True when convergence has been reached.
        """
        log.debug("Entering is_converged()")
        #print "max_dm_dt=%s < %s=stopping_dm_dt ? %s" \
        #      % (max_dm_dt, stopping_dm_dt, max_dm_dt < stopping_dm_dt)
        if max_dm_dt < stopping_dm_dt:
            self.countdown -= 1
            if self.countdown < 1:
                log.debug("is_converged(): True")
                # Since we are converging, we don't want to interfere
                # and change the tolerances now! We therefore return
                # new_tol_factor = None
                self._reset()
                return True, None

            log.debug("is_converged(): Trueish (%d times converged but need "
                      "%d times to stop)" % (self.countdown, self.countdown))
            return False, None

        self.countdown = self.countdown_initial
        log.debug("is_converged(): False")

        convergence_quality = self._quality(max_dm_dt)
        self.log.append((step, max_dm_dt, stopping_dm_dt, convergence_quality))

        if convergence_quality == None:
            return False, None

        else:
            new_tol_factor = None
            if convergence_quality == 1.0:
                new_tol_factor = tol_factor*self.worsen_factor
            elif convergence_quality < 0.5:
                new_tol_factor = tol_factor*self.improve_factor
            if new_tol_factor != None:
                new_tol_factor = min(self.max_factor,
                                     max(self.min_factor, new_tol_factor))
                if new_tol_factor == tol_factor: new_tol_factor = None
        return False, new_tol_factor

    def get_log(self):
        l = self.log.get_list()
        s = "# Convergence log\n"
        s += "# step, ax dm/dt, stoppping dm/dt, conv. quality\n"
        for step, max_dm_dt, stopping_dm_dt, conv_quality in l:
            s += "%s %s %s %s\n" % (step, max_dm_dt, stopping_dm_dt,
                                    conv_quality)
        return s