Changeset 536

Timestamp:

07/29/12 22:10:39 (4 years ago)

Author:

mmckerns

Message:

added guard against null and empty logs in munge; added isNull;
allow non-boolean returns from termination to provide information for restarts

Location:

mystic/mystic

Files:

• differential_evolution.py (modified) (2 diffs)
• monitors.py (modified) (2 diffs)
• munge.py (modified) (6 diffs)
• scipy_optimize.py (modified) (2 diffs)
• termination.py (modified) (15 diffs)
• tools.py (modified) (1 diff)

mystic/mystic/differential_evolution.py

@@ -286 +286 @@
 
         self.energy_history.append(self.bestEnergy)
+        termination(self) #XXX: initialize termination conditions, if needed
         self._stepmon(self.bestSolution[:], self.bestEnergy, id)
-        self.generations = 0  #XXX: above currently *not* counted as an iteration
+        self.generations = 0 #XXX: above currently *not* counted as an iteration
         if callback is not None:
             callback(self.bestSolution)
@@ -493 +494 @@
 
         self.energy_history.append(self.bestEnergy)
+        termination(self) #XXX: initialize termination conditions, if needed
        #FIXME: StepMonitor works for 'pp'?
         self._stepmon(self.bestSolution[:], self.bestEnergy, id)

mystic/mystic/monitors.py

@@ -75 +75 @@
     def __setattr__(self, name, value): return self
     def __delattr__(self, name): return self
+    def __len__(self): return  #XXX ?
 
 class Monitor(object):
@@ -99 +100 @@
         self._info = []
        #self._all = all
+
+    def __len__(self):
+        return len(self.x)
 
     def info(self, message):

mystic/mystic/munge.py

@@ -1 +1 @@
 from mystic.tools import list_or_tuple_or_ndarray as sequence
+from mystic.tools import isNull
 
 # logfile reader
@@ -46 +47 @@
 
 def raw_to_converge(steps, energy):
-  if not sequence(steps[0][0]):
-    steps = [[step] for step in steps]  # needed when steps = [1,2,3,...]
-  steps = [zip(*step) for step in steps] # also can be used to revert 'steps'
+  if len(steps) > 0:
+    if not sequence(steps[0][0]):
+      steps = [[step] for step in steps]  # needed when steps = [1,2,3,...]
+    steps = [zip(*step) for step in steps] # also can be used to revert 'steps'
   return steps, energy
 
@@ -58 +60 @@
 
 def write_raw_file(mon,log_file='paramlog.py',**kwds):
+  if isNull(mon): return  #XXX: throw error? warning? ???
   steps, energy = read_monitor(mon)
   f = open(log_file,'w')
@@ -69 +72 @@
 
 def write_support_file(mon,log_file='paramlog.py'):
+  if isNull(mon): return  #XXX: throw error? warning? ???
   monitor = write_monitor( *raw_to_support( *read_monitor(mon) ) )
   write_raw_file(monitor,log_file,header="written in 'support' format")
@@ -74 +78 @@
 
 def write_converge_file(mon,log_file='paramlog.py'):
+  if isNull(mon): return  #XXX: throw error? warning? ???
   monitor = write_monitor( *raw_to_converge( *read_monitor(mon) ) )
   write_raw_file(monitor,log_file,header="written in 'converge' format")
@@ -131 +136 @@
 
 def __orig_write_support_file(mon,log_file='paramlog.py'):
+  if isNull(mon): return  #XXX: throw error? warning? ???
   steps, energy = read_monitor(mon)
   log = []
-  for p in range(len(steps[0])):
-    q = []
-    for s in range(len(steps)):
-      q.append(steps[s][p])
-    log.append(q)  
+  if len(steps) > 0:
+    for p in range(len(steps[0])):
+      q = []
+      for s in range(len(steps)):
+        q.append(steps[s][p])
+      log.append(q)  
   monitor = write_monitor(log, energy)
   write_raw_file(monitor,log_file)

mystic/mystic/scipy_optimize.py

@@ -215 +215 @@
             allvecs = [sim[0]]
         fsim[0] = func(x0)
+        termination(self) #XXX: initialize termination conditions, if needed
         self._stepmon(sim[0], fsim[0], id) # sim = all; "best" is sim[0]
 
@@ -586 +587 @@
         self.popEnergy[0] = fval  #XXX: pointless?
         self.energy_history.append(self.bestEnergy)
+        termination(self) #XXX: initialize termination conditions, if needed
         self._stepmon(x, fval, id) # get initial values
 

mystic/mystic/termination.py

@@ -7 +7 @@
 import numpy
 from numpy import absolute
+from __builtin__ import bool as _bool
 abs = absolute
 Inf = numpy.Inf
+null = ""
 
 # a module level singleton.
@@ -18 +20 @@
 
 cost[-1] <= tolerance"""
-    def _VTR(inst):
+    doc = "VTR with %s" % {'tolerance':tolerance, 'target':target}
+    def _VTR(inst, bool=True):
+         if bool: bool = _bool
+         else: bool = lambda x:x
          hist = inst.energy_history
-         return abs(hist[-1] - target) <= tolerance
+         if not len(hist): return bool(null)
+         if abs(hist[-1] - target) <= tolerance: msg = doc
+         else: msg = null
+         return bool(msg)
+   #_VTR.__doc__ = "%s(**%s)" % tuple(doc.split(" with "))
+    _VTR.__doc__ = doc
     return _VTR
 
@@ -27 +37 @@
 
 cost[-g] - cost[-1] <= tolerance, where g=generations"""
+    doc = "ChangeOverGeneration with %s" % {'tolerance':tolerance,
+                                            'generations':generations}
     def _ChangeOverGeneration(inst):
          hist = inst.energy_history
          lg = len(hist)
-         if lg <= generations: return False
-         return (hist[-generations]-hist[-1]) <= tolerance
+         if lg <= generations: return ""
+         if (hist[-generations]-hist[-1]) <= tolerance: return doc
+         return ""
+    _ChangeOverGeneration.__doc__ = doc
     return _ChangeOverGeneration
 
@@ -39 +53 @@
 (cost[-g] - cost[-1]) /  0.5*(abs(cost[-g]) + abs(cost[-1])) <= tolerance"""
     eta = 1e-20
+    doc = "NormalizedChangeOverGeneration with %s" % {'tolerance':tolerance,
+                                                      'generations':generations}
     def _NormalizedChangeOverGeneration(inst):
          hist = inst.energy_history
          lg = len(hist)
-         if lg <= generations: return False
+         if lg <= generations: return ""
          diff = tolerance*(abs(hist[-generations])+abs(hist[-1])) + eta
-         return 2.0*(hist[-generations]-hist[-1]) <= diff
+         if 2.0*(hist[-generations]-hist[-1]) <= diff: return doc
+         return ""
+    _NormalizedChangeOverGeneration.__doc__ = doc
     return _NormalizedChangeOverGeneration
               
@@ -52 +70 @@
 abs(xi-x0) <= xtol & abs(fi-f0) <= ftol, where x=params & f=cost"""
     #NOTE: this termination expects nPop > 1
+    doc = "CandidateRelativeTolerance with %s" % {'xtol':xtol, 'ftol':ftol}
     def _CandidateRelativeTolerance(inst):
          sim = numpy.array(inst.population)
          fsim = numpy.array(inst.popEnergy)
          if not len(fsim[1:]):
-             print "Warning: Invalid termination condition (nPop < 2)"
-             return True
+             warn = "Warning: Invalid termination condition (nPop < 2)"
+             print warn
+             return warn
          #   raise ValueError, "Invalid termination condition (nPop < 2)"
          #FIXME: abs(inf - inf) will raise a warning...
@@ -64 +84 @@
          answer = answer and max(abs(fsim[0]-fsim[1:])) <= ftol
          numpy.seterr(invalid=errdict['invalid']) #FIXME: turn on warnings
-         return answer
+         if answer: return doc
+         return ""
+    _CandidateRelativeTolerance.__doc__ = doc
     return _CandidateRelativeTolerance
 
@@ -71 +93 @@
 
 sum(abs(last_params - current_params)) <= tolerance"""
+    doc = "SolutionImprovement with %s" % {'tolerance':tolerance}
     def _SolutionImprovement(inst):
         best = numpy.array(inst.bestSolution)
@@ -76 +99 @@
         update = best - trial #XXX: if inf - inf ?
         answer = numpy.add.reduce(abs(update)) <= tolerance
-        return answer
+        if answer: return doc
+        return ""
+    _SolutionImprovement.__doc__ = doc
     return _SolutionImprovement
 
@@ -87 +112 @@
     #  original --> if generations: then return cost[-g] - cost[-1] < 0
     #           --> else: return fval != 0 and abs((best - fval)/fval) < tol
+    doc = "NormalizedCostTarget with %s" % {'fval':fval, 'tolerance':tolerance,
+                                            'generations':generations}
     def _NormalizedCostTarget(inst):
          if generations and fval == None:
@@ -92 +119 @@
              lg = len(hist)
              #XXX: throws error when hist is shorter than generations ?
-             return lg > generations and (hist[-generations]-hist[-1]) <= 0
-         if not generations and fval == None: return True
-         return abs(inst.bestEnergy-fval) <= abs(tolerance * fval)
+             if lg > generations and (hist[-generations]-hist[-1]) <= 0:
+                 return doc
+             return ""
+         if not generations and fval == None: return doc
+         if abs(inst.bestEnergy-fval) <= abs(tolerance * fval): return doc
+         return ""
+    _NormalizedCostTarget.__doc__ = doc
     return _NormalizedCostTarget
 
@@ -103 +134 @@
 
 cost[-g] - cost[-1] <= gtol, where g=generations *or* cost[-1] <= ftol."""
+    doc = "VTRChangeOverGeneration with %s" % {'ftol':ftol, 'gtol':gtol,
+                                               'generations':generations,
+                                               'target':target}
     def _VTRChangeOverGeneration(inst):
          hist = inst.energy_history
          lg = len(hist)
          #XXX: throws error when hist is shorter than generations ?
-         return (lg > generations and (hist[-generations]-hist[-1]) <= gtol)\
-                or ( abs(hist[-1] - target) <= ftol )
+         if (lg > generations and (hist[-generations]-hist[-1]) <= gtol)\
+                or ( abs(hist[-1] - target) <= ftol ): return doc
+         return ""
+    _VTRChangeOverGeneration.__doc__ = doc
     return _VTRChangeOverGeneration
 
@@ -115 +151 @@
 
 abs(params - params[0]) <= tolerance"""
+    doc = "PopulationSpread with %s" % {'tolerance':tolerance}
     def _PopulationSpread(inst):
          sim = numpy.array(inst.population)
@@ -120 +157 @@
          #    print "Warning: Invalid termination condition (nPop < 2)"
          #    return True
-         return numpy.all(abs(sim - sim[0]) <= abs(tolerance * sim[0]))
+         if numpy.all(abs(sim - sim[0]) <= abs(tolerance * sim[0])): return doc
+         return ""
+    _PopulationSpread.__doc__ = doc
     return _PopulationSpread
 
@@ -127 +166 @@
 
 sum( abs(gradient)**norm )**(1.0/norm) <= tolerance"""
+    doc = "GradientNormTolerance with %s" % {'tolerance':tolerance, 'norm':norm}
     def _GradientNormTolerance(inst):
         try:
             gfk = inst.gfk #XXX: need to ensure that gfk is an array ?
         except:
-            print "Warning: Invalid termination condition (no gradient)"
-            return True
+            warn = "Warning: Invalid termination condition (no gradient)"
+            print warn
+            return warn
         if norm == Inf:
             gnorm = numpy.amax(abs(gfk))
@@ -141 +182 @@
            #XXX: as norm < -large, gnorm approaches amin(abs(gfk)) --> then -inf
             gnorm = numpy.sum(abs(gfk)**norm,axis=0)**(1.0/norm)
-        return gnorm <= tolerance
+        if gnorm <= tolerance: return doc
+        return ""
+    _GradientNormTolerance.__doc__ = doc
     return _GradientNormTolerance
 

mystic/mystic/tools.py

@@ -251 +251 @@
 from monitors import Null
 
+def isNull(mon):
+    if isinstance(mon, Null): # is Null()
+        return True
+    if mon == Null:  # is Null
+        return True
+    return False
+
 
 if __name__=='__main__':