Moved hv.py to pyhv and made the import transparent in indicator.

Added debug info in cma_mo

--HG--
branch : dev
rename : deap/tools/hv.py => deap/tools/pyhv.py

deap/tools/indicator.py

@@ -16,7 +16,12 @@
 
 import numpy
 
-import deap.tools.hv
+try:
+    # try importing the C version
+    from . import hv as hv
+except ImportError:
+    # fallback on python version
+    from . import pyhv as hv
 
 def hypervolume(front, ref=None):
     """Returns the index of the individual with the least the hypervolume
@@ -29,7 +34,7 @@ def hypervolume(front, ref=None):
         ref = numpy.max(wobj, axis=0) + 1
 
     def contribution(i):
-        return deap.tools.hv.hypervolume(numpy.concatenate((wobj[:i], wobj[i+1:])), ref)
+        return hv.hypervolume(numpy.concatenate((wobj[:i], wobj[i+1:])), ref)
 
     # TODO: Parallelize this?
     contrib_value = map(contribution, range(len(front)))

deap/tools/hv.py → deap/tools/pyhv.py

@@ -19,59 +19,13 @@
 __author__ = "Simon Wessing"
 # Wrapped by Francois-Michel De Rainville
 
-def hypervolume_kmax(front, k, ref=None):
-    """Find the *k* individuals index contributing the most to
-    the hypervolume among *front*.
-    """
-    if k >= len(front):
-        return range(len(front))
-
-    # Must use wvalues * -1 since _HyperVolume use implicit minimization
-    wobj = numpy.array([ind.fitness.wvalues for ind in front]) * -1
-    if ref is None:
-        ref = numpy.max(wobj, axis=0) + 1
-
-    # print "ref", ref
-
-    hv = _HyperVolume(ref)
-    indices = numpy.arange(0, len(front))
-    contrib = numpy.zeros(len(front))
-
-    # import matplotlib.pyplot as plt
-    # plt.scatter(wobj[:, 0], wobj[:, 1])
-    # plt.show()
-
-    for i in range(len(front) - k):
-        for j in indices:
-            indices_j = indices[numpy.where(indices != j)]
-            s_a_j = hv.compute(wobj[indices_j])
-            contrib[j] = s_a_j
-
-        # Select randomly from equaly contributing
-        ## Retreive the indices
-        least_contributers = numpy.flatnonzero(numpy.isclose(contrib, contrib.max()))
-        idx = numpy.random.choice(least_contributers)
-        indices = indices[numpy.where(indices != idx)]
-        contrib[idx] = 0
-        # print "====="
-        # print s_a
-        # print contrib
-        # print indices, idx, contrib.min()
-
-        # plt.scatter(wobj[:, 0], wobj[:, 1], c="r")
-        # plt.scatter(wobj[indices, 0], wobj[indices, 1], c="b")
-        # plt.show()
-    return indices
-
-def hypervolume(population, ref=None):
-    """Compute the absolute hypervolume of a *population*."""
-    # Must use (wvalues * -1) since _HyperVolume use implicit minimization
-    wobj = numpy.array([-ind.fitness.wvalues for ind in population]) * -1
-    if ref is None:
-        ref = numpy.max(wobj, axis=0) + 1
 
+def hypervolume(pointset, ref):
+    """Compute the absolute hypervolume of a *pointset* according to the
+    reference point *ref*.
+    """
     hv = _HyperVolume(ref)
-    return hv.compute(wobj)
+    return hv.compute(pointset)
 
 
 class _HyperVolume:

examples/es/cma_mo.py

@@ -22,6 +22,8 @@
 from deap import creator
 from deap import tools
 
+import cma_mo_debug
+
 # Problem size
 N = 30
 
@@ -65,14 +67,16 @@ def main():
     distances = list()
 
     MU, LAMBDA = 100, 100
-    NGEN = 500
+    NGEN = 1
     verbose = False
 
     # The MO-CMA-ES algorithm takes a full population as argument
-    population = [creator.Individual(x) for x in (numpy.random.uniform(0, 1, (MU + LAMBDA, N)))]
+    # population = [creator.Individual(x) for x in (numpy.random.uniform(0, 1, (MU + LAMBDA, N)))]
     # init = numpy.zeros((MU, N))
     # init[:, 0] = numpy.linspace(0, 1, 100)
     # population = [creator.Individual(x) for x in init]
+    population = [creator.Individual(z) for z in cma_mo_debug.parents]
+
     for ind in population:
         ind.fitness.values = toolbox.evaluate(ind)
 
@@ -95,7 +99,11 @@ def main():
 
     for gen in range(NGEN):
         # Generate a new population
-        population = toolbox.generate()
+        # population = toolbox.generate()
+        population = [creator.Individual(z) for z in cma_mo_debug.offspring]
+        for i, ind in enumerate(population):
+            ind._ps = "o", i
+
         # Evaluate the individuals
         fitnesses = toolbox.map(toolbox.evaluate, population)
         for ind, fit in zip(population, fitnesses):
@@ -127,24 +135,24 @@ def main():
 
         print("success", strategy.success_count)
 
-    import matplotlib.pyplot as plt
-    # from mpl_toolkits.mplot3d import Axes3D
+    # import matplotlib.pyplot as plt
+    # # from mpl_toolkits.mplot3d import Axes3D
 
-    valid_front = numpy.array([ind.fitness.values for ind in population if valid(ind)])
-    invalid_front = numpy.array([ind.fitness.values for ind in population if not valid(ind)])
+    # valid_front = numpy.array([ind.fitness.values for ind in population if valid(ind)])
+    # invalid_front = numpy.array([ind.fitness.values for ind in population if not valid(ind)])
 
-    fig = plt.figure()
-    # ax = fig.add_subplot(111, projection='3d')
-    # ax.scatter(front[:,0], front[:,1], front[:,2], c="b")
+    # fig = plt.figure()
+    # # ax = fig.add_subplot(111, projection='3d')
+    # # ax.scatter(front[:,0], front[:,1], front[:,2], c="b")
 
-    plt.scatter(valid_front[:,0], valid_front[:,1], c="g")
-    plt.scatter(invalid_front[:,0], invalid_front[:,1], c="r")
+    # plt.scatter(valid_front[:,0], valid_front[:,1], c="g")
+    # plt.scatter(invalid_front[:,0], invalid_front[:,1], c="r")
 
-    plt.figure()
-    sigmas = numpy.mean(sigmas, axis=0)
-    plt.plot(sigmas)
+    # plt.figure()
+    # sigmas = numpy.mean(sigmas, axis=0)
+    # plt.plot(sigmas)
 
-    plt.show()
+    # plt.show()
 
     return halloffame