Changeset 177

Timestamp:

08/10/09 13:37:47 (7 years ago)

Author:

altafang

Message:

Fixing small things in snobfit and test_snobfit

Location:

branches/alta/mystic-0.1a2

Files:

• examples/test_snobfit.py (modified) (9 diffs)
• mystic/snobfit_solver.py (modified) (10 diffs)

branches/alta/mystic-0.1a2/examples/test_snobfit.py

@@ -7 +7 @@
 Demonstrates:
     - minimal solver interface for snobfit
+    - Use of a mystic solver without just the minimal interface
+    - Use of SnobfitSolver and SnobfitSoftSolver
 
 These tests are adapted from the snobfit.py file.
@@ -13 +15 @@
 from mystic.snobfit_solver import snobfit
 from mystic.softConstraint import SoftConstraint, SoftConstraints
+from mystic.tools import random_seed
 import numpy
 
@@ -25 +28 @@
     return f
 
+def hsf18(x):
+    f = 0.01*x[0]**2 + x[1]**2
+    return f
+
 iter = 0
 # plot the parameter trajectories
@@ -33 +40 @@
     iter += 1
     return
-
-def hsf18(x):
-    f = 0.01*x[0]**2 + x[1]**2
-    return f
 
 #---------------------------------------------------------
@@ -60 +63 @@
     v = -u
     xglob = numpy.array([1, 1])
-    results = snobfit(ros, x0, (u, v), seed = 1, callback=print_params,\
-                      disp=1,full_output=1)
+    random_seed(1)
+    results = snobfit(ros, x0, (u, v), callback=print_params,\
+                      disp=1,full_output=1, retuct=True)
     print('optimize:', results)
 
@@ -85 +89 @@
     s.add(  SoftConstraint( F=hsc18_2, Flo=0, Fhi=numpy.inf, sigma=1.25) )
 
+    random_seed(1)
     results = snobfit(hsf18, x0, (u, v),
-                                  constraint=s, full_output=1,
-                                  retall=1, 
-                                  disp=1, seed = 1
-                                  )
+                                  constraint=s, 
+                                  #retall=1, callback=print_params,
+                                  disp=1, full_output=1)
     print results
 
@@ -96 +100 @@
 #         Rosenbrock function.
 
-def test_4():
+def test_ros2():
+    """Minimize the Rosenbrock function."""
+
     from mystic.tools import Sow, random_seed
+    from mystic.termination import VTRChangeOverGenerations
+    from snobfit_solver import SnobfitSolver
+
+    # Create the step and evaluation monitors, which capture progress 
     stepmon = Sow()
     evalmon = Sow()
-    from mystic.termination import SnobfitTermination
-    from mystic.snobfit_solver import SnobfitSolver
 
+    # The initial guess x0 and bounds u,v
     x0 = numpy.array([2, 3])
     u = -5.12*numpy.ones(2)
@@ -108 +117 @@
 
     random_seed(1)
-    solver = SnobfitSolver(len(x0))
+
+    # Initialize the solver
+    solver = SnobfitSolver(len(x0),7)
     solver.SetStrictRanges(u, v)
     solver.SetInitialPoints(x0)
     solver.enable_signal_handler()
 
-    solver.Solve(ros,termination=SnobfitTermination(),\
+    # Call the main Solve() function
+    solver.Solve(ros,termination=VTRChangeOverGenerations(1e-6, 0, 250),\
                  EvaluationMonitor=evalmon,StepMonitor=stepmon)
-    solution = solver.Solution()
 
+    # Get solution
     x = solver.bestSolution
+
+    # Get the function value at the solution
     fval = solver.bestEnergy
+
+    # Get the number of function evaluations and iterations
     fcalls = len(evalmon.x)
     iterations = len(stepmon.x)
 
-    print ('optimize:', x, fval, iterations, fcalls)
+    # Output the results
+    print "Solution: ", x
+    print "Function value at solution: ", fval
+    print "Iterations: ", iterations
+    print "Function calls: ", fcalls
 
 #------------------------------------------------------------------
@@ -128 +148 @@
 if __name__ == "__main__":
     #test_1()
-    test_ros()
-    #test_hsf18() 
-    #test_4()
+    test_ros()       # Uses callback
+    #test_hsf18()     # Uses SnobfitSoftSolver 
+    #test_ros2()

branches/alta/mystic-0.1a2/mystic/snobfit_solver.py

@@ -83 +83 @@
     """Snobfit optimization. """
 
-    def __init__(self, dim, **kwds):
+    def __init__(self, dim, NP=None):
         """
 Required input: 
@@ -94 +94 @@
                 than parameters. Default is dn = 5 and NP adjusted accordingly.
         """
-        if kwds.has_key("NP"):
+        if NP:
             NP = NP
         else:
@@ -155 +155 @@
             self.func = wrap_bounds(self.func, self._strictMin, self._strictMax)
 
-            # Fix for when bounds were set after initial points: reinitialize
+            # Re-initialize population if bounds were set after initial points. 
             # This may give different results from if SetStrictRanges was called first
             x0 = self.population[0]
@@ -167 +167 @@
         x0 = self.population[0]
         self.x = self.population
-
-        self.fglob = 0
 
         # The number of safeguarded nearest neighbors
@@ -270 +268 @@
         return 
 
-    # Overwrite abstract_solver's SetRandomInitialPoints()
+    # Overwrite AbstractSolver's SetRandomInitialPoints()
     
-    # If SetStrictRange was called first, use those bounds.
+    # If SetStrictRanges was called first, use those bounds.
     # If SetInitialPoints was called first, temporarily use defaults,
-    # and Solve() will re-initialize with user-specified bounds
+    # and Solve() will re-initialize with the bounds from SetStrictRanges
     def SetRandomInitialPoints(self,min=None,max=None):
         if self._useStrictRange:
@@ -287 +285 @@
             if min is None: min = [-1e3]*self.nDim  
             if max is None: max = [1e3]*self.nDim
-        self.population = self._setInitRecommendedEvalPoints(min,max) 
-        print self.population
-   # def SetInitialPoints(self,x0):
-   #     self.population = self._setInitRecommendedEvalPoints(self._strictMin,self._strictMax)
-   #     self.population[0] = x0
+        self.population = self._setInitRecommendedEvalPoints(min,max)
         
 ############################################################################
 # Helper methods for SnobfitSolver
 
-    def _setInitRecommendedEvalPoints(self,min,max):
+    def _setInitRecommendedEvalPoints(self, min, max):
         """
         Snobfit's initialization of the population
@@ -2367 +2361 @@
             self.func = wrap_bounds(self.func, self._strictMin, self._strictMax)
 
-            # Fix for when bounds were set after initial points. 
+            # Re-initialize population if bounds were set after initial points. 
             x0 = self.population[0]
             self.SetInitialPoints(x0)
@@ -2681 +2675 @@
         1 : 'Maximum number of function evaluations.'
         2 : 'Maximum number of iterations.'
-    allvecs -- list - a list of solutions at each iteration
+    allvecs -- list - Returned if retall is true. 
+                      A list of solutions at each iteration.
     uncertainty -- list - Returned if retuct is True. The uncertainty of 
                    the fitting parameters.
@@ -2687 +2682 @@
 
     from mystic.tools import Sow
-    from mystic.termination import FglobTermination, VTRChangeOverGenerations
-
     stepmon = Sow()
     evalmon = Sow()
 
+    # Use SnobfitSoftSolver if there are SoftConstraints.
+    # Pass npop to the solver if it was specified. If it was not, 
+    # let the solver use the default.
     if constraint is None:
         if npop:
@@ -2704 +2700 @@
 
     if fglob != 0:
-        termination = FglobTermination(fglob=fglob, tolerance=rtol, generations=nstop)
+        from mystic.termination import FglobTermination
+        termination = FglobTermination(fglob, rtol, nstop)
     else:
-        termination = VTRChangeOverGenerations(costtolerance=ftol, gentolerance=0,\
-                                               generations=nstop)
+        from mystic.termination import VTRChangeOverGenerations
+        termination = VTRChangeOverGenerations(ftol, 0, nstop)
 
     # Bounds are required