Changeset 81

Timestamp:

01/23/09 13:03:25 (7 years ago)

Author:

mmckerns

Message:

make nelder_mead's SimplexMonitor? fit StepMonitor? interface

Files:

• python/nelder_mead.py (modified) (6 diffs)
• python/tools.py (modified) (1 diff)
• tests/sam_corana2.py (modified) (1 diff)
• tests/sam_mogi.py (modified) (3 diffs)
• tests/sam_rosenbrock.py (modified) (2 diffs)
• tests/sam_zimmermann.py (modified) (2 diffs)
• tests/test_mogi.py (modified) (1 diff)
• tests/test_zimmermann.py (modified) (1 diff)

python/nelder_mead.py

@@ -12 +12 @@
                      Takes two additional input args
                         -- EvaluationMonitor
-                        -- SimplexMonitor
+                        -- StepMonitor
                      
 """
@@ -30 +30 @@
 
 def fmin(func, x0, args=(), xtol=1e-4, ftol=1e-4, maxiter=None, maxfun=None,
-         full_output=0, disp=1, retall=0, EvaluationMonitor=Null, SimplexMonitor=Null):
+         full_output=0, disp=1, retall=0, EvaluationMonitor=Null, StepMonitor=Null):
     """Minimize a function using the downhill simplex algorithm.
 
@@ -70 +70 @@
       EvaluationMonitor -- a callable object that will be passed x, fval
            whenever the cost function is evaluated.
+      StepMonitor -- a callable object that will be passed x, fval
+           after the end of a simplex iteration.
 
       """
@@ -119 +121 @@
 
     while (fcalls[0] < maxfun and iterations < maxiter):
-        SimplexMonitor(sim, fsim)
+        StepMonitor(sim, fsim) # get all values; "best" is sim[0]
         if (max(numpy.ravel(abs(sim[1:]-sim[0]))) <= xtol \
             and max(abs(fsim[0]-fsim[1:])) <= ftol):
@@ -219 +221 @@
     print "==================="
     start = time.time()
-    x = fmin(rosen,x0)
+    from tools import VerboseSow
+    stepmon = VerboseSow(10)
+    x = fmin(rosen,x0,StepMonitor=stepmon)
     print x
     times.append(time.time() - start)
@@ -225 +229 @@
 
     for k in range(len(algor)):
-        print algor[k], "\t -- ", times[k]
+        print algor[k], "\t -- took", times[k]
 
 if __name__ == "__main__":

python/tools.py

@@ -148 +148 @@
         return
     def __call__(self, x, y):
+        from numpy import ndarray
         Sow.__call__(self, x, y)
         self._step += 1
+        if isinstance(y,(list,ndarray)):
+            y = y[0] #XXX: get the "best" fit... which should be in y[0]
+        if isinstance(x[0],(list,ndarray)): #XXX: x should always be iterable
+            x = x[0] #XXX: get the "best" fit... which should be in x[0]
         if int(self._step % self._yinterval) == 0:
+           #print "Generation %d has best Chi-Squared: %s" % (self._step, y)
             print "Generation %d has best Chi-Squared: %f" % (self._step, y)
         if int(self._step % self._xinterval) == 0:
-            print "Generation %d has bet fit parameters: %r" % (self._step, x)
+            print "Generation %d has bet fit parameters: %s" % (self._step, x)
         return
     pass

tests/sam_corana2.py

@@ -42 +42 @@
 def run_once():
     simplex = Sow()
-    sol = fmin(Corana2, [random.uniform(0,2), random.uniform(0,2)], retall = True, SimplexMonitor = simplex)
+    sol = fmin(Corana2, [random.uniform(0,2), random.uniform(0,2)], retall = True, StepMonitor = simplex)
     
     for x in simplex.x:

tests/sam_mogi.py

@@ -4 +4 @@
 See test_mogi.py
 
-This uses Nelder-Mead and SimplexMonitor
+This uses Nelder-Mead and StepMonitor
 """
 
@@ -61 +61 @@
     z1 = z0*random.uniform(0.5,1.5)
     v1 = v0*random.uniform(0.5,1.5)
-    sol = fmin(cost_function, [random.uniform(x0-40,x0+40), random.uniform(y0-40,y0+40), z1, v1], retall = True, SimplexMonitor = simplex)
+    sol = fmin(cost_function, [random.uniform(x0-40,x0+40), random.uniform(y0-40,y0+40), z1, v1], retall = True, StepMonitor = simplex)
     print sol[0]
     
@@ -74 +74 @@
     y1 = y0*random.uniform(0.5,1.5)
     z1 = z0*random.uniform(0.5,1.5)
-    sol = fmin(cost_function, [random.uniform(x0-40,x0+40), y1, z1, random.uniform(v0-0.1,v0+0.1)], retall = True, SimplexMonitor = simplex)
+    sol = fmin(cost_function, [random.uniform(x0-40,x0+40), y1, z1, random.uniform(v0-0.1,v0+0.1)], retall = True, StepMonitor = simplex)
     print sol[0]
     

tests/sam_rosenbrock.py

@@ -6 +6 @@
 This one uses Nelder-Mead plus matlab viz.
 
-It uses the SimplexMonitor option to track all simplices generated during 
+It uses the StepMonitor option to track all simplices generated during 
 the search.
 """
@@ -40 +40 @@
 def run_once(x0,x1):
     simplex = Sow()
-    sol = fmin(rosen, [x0, x1], retall = True, SimplexMonitor = simplex)
+    sol = fmin(rosen, [x0, x1], retall = True, StepMonitor = simplex)
     
     for x in simplex.x:

tests/sam_zimmermann.py

@@ -6 +6 @@
 This one uses Nelder-Mead plus matlab viz.
 
-It uses the SimplexMonitor option to track all simplices generated during 
+It uses the StepMonitor option to track all simplices generated during 
 the search.
 """
@@ -39 +39 @@
 def run_once():
     simplex = Sow()
-    sol = fmin(CostFunction, [random.uniform(0,7), random.uniform(0,7)], retall = True, SimplexMonitor = simplex)
+    sol = fmin(CostFunction, [random.uniform(0,7), random.uniform(0,7)], retall = True, StepMonitor = simplex)
     
     for x in simplex.x:

tests/test_mogi.py

@@ -125 +125 @@
     #
     simplex, esow = Sow(), Sow()
-    sol = fmin(cost_function, point,  EvaluationMonitor = esow, retall = True, SimplexMonitor = simplex)
+    sol = fmin(cost_function, point,  EvaluationMonitor = esow, retall = True, StepMonitor = simplex)
     print "simplex solution: ", sol[1][-1]
     #

tests/test_zimmermann.py

@@ -67 +67 @@
     simplex = Sow()
     esow = Sow()
-    sol = fmin(CostFunction, [random.uniform(0,5) for j in range(ND)], EvaluationMonitor = esow, retall = True, SimplexMonitor = simplex)
+    sol = fmin(CostFunction, [random.uniform(0,5) for j in range(ND)], EvaluationMonitor = esow, retall = True, StepMonitor = simplex)
     #print "solution: ", sol