Bugfixes related to VL-BFGS and SteepestDescent

demos/wiener_filter_hamiltonian.py

+
+from nifty import *
+import plotly.offline as pl
+import plotly.graph_objs as go
+
+from mpi4py import MPI
+comm = MPI.COMM_WORLD
+rank = comm.rank
+
+
+if __name__ == "__main__":
+
+    distribution_strategy = 'fftw'
+
+    s_space = RGSpace([512, 512], dtype=np.float)
+    fft = FFTOperator(s_space)
+    h_space = fft.target[0]
+    p_space = PowerSpace(h_space, distribution_strategy=distribution_strategy)
+
+    pow_spec = (lambda k: (42 / (k + 1) ** 3))
+
+    S = create_power_operator(h_space, power_spectrum=pow_spec,
+                              distribution_strategy=distribution_strategy)
+
+    sp = Field(p_space, val=lambda z: pow_spec(z)**(1./2),
+               distribution_strategy=distribution_strategy)
+    sh = sp.power_synthesize(real_signal=True)
+    ss = fft.inverse_times(sh)
+
+    R = SmoothingOperator(s_space, sigma=0.01)
+
+#    R = DiagonalOperator(s_space, diagonal=1.)
+#    R._diagonal.val[200:400, 200:400] = 0
+
+    signal_to_noise = 1
+    N = DiagonalOperator(s_space, diagonal=ss.var()/signal_to_noise, bare=True)
+    n = Field.from_random(domain=s_space,
+                          random_type='normal',
+                          std=ss.std()/np.sqrt(signal_to_noise),
+                          mean=0)
+    #n.val.data.imag[:] = 0
+
+    d = R(ss) + n
+    j = R.adjoint_times(N.inverse_times(d))
+    D = PropagatorOperator(S=S, N=N, R=R)
+
+    def energy(x):
+        DIx = D.inverse_times(x)
+        H = 0.5 * DIx.dot(x) - j.dot(x)
+        return H.real
+
+    def gradient(x):
+        DIx = D.inverse_times(x)
+        g = DIx - j
+        return_g = g.copy_empty(dtype=np.float)
+        return_g.val = g.val.real
+        return return_g
+
+    def distance_measure(x, fgrad, iteration):
+        print (iteration, ((x-ss).norm()/ss.norm()).real)
+
+    minimizer = SteepestDescent(convergence_tolerance=0,
+                                iteration_limit=50,
+                                callback=distance_measure)
+    minimizer = VL_BFGS(convergence_tolerance=0,
+                        iteration_limit=50,
+                        callback=distance_measure,
+                        max_history_length=5)
+
+
+    m0 = Field(s_space, val=1)
+
+    (m, convergence) = minimizer(m0, energy, gradient)
+
+
+    grad = gradient(m)
+
+    d_data = d.val.get_full_data().real
+    if rank == 0:
+        pl.plot([go.Heatmap(z=d_data)], filename='data.html')
+
+
+    ss_data = ss.val.get_full_data().real
+    if rank == 0:
+        pl.plot([go.Heatmap(z=ss_data)], filename='ss.html')
+
+    sh_data = sh.val.get_full_data().real
+    if rank == 0:
+        pl.plot([go.Heatmap(z=sh_data)], filename='sh.html')
+
+    j_data = j.val.get_full_data().real
+    if rank == 0:
+        pl.plot([go.Heatmap(z=j_data)], filename='j.html')
+
+    jabs_data = np.abs(j.val.get_full_data())
+    jphase_data = np.angle(j.val.get_full_data())
+    if rank == 0:
+        pl.plot([go.Heatmap(z=jabs_data)], filename='j_abs.html')
+        pl.plot([go.Heatmap(z=jphase_data)], filename='j_phase.html')
+
+    m_data = m.val.get_full_data().real
+    if rank == 0:
+        pl.plot([go.Heatmap(z=m_data)], filename='map.html')
+
+    grad_data = grad.val.get_full_data().real
+    if rank == 0:
+        pl.plot([go.Heatmap(z=grad_data)], filename='grad.html')

nifty/minimization/bare_dot.py

-# -*- coding: utf-8 -*-
-
-import numpy as np
-
-
-def bare_dot(a, b):
-    try:
-        return a.dot(b, bare=True)
-    except(AttributeError, TypeError):
-        pass
-
-    try:
-        return a.vdot(b)
-    except(AttributeError):
-        pass
-
-    return np.vdot(a, b)

nifty/minimization/line_searching/line_search.py

@@ -2,8 +2,6 @@ import abc
 
 from keepers import Loggable
 
-from ..bare_dot import bare_dot
-
 
 class LineSearch(object, Loggable):
     """
@@ -94,7 +92,7 @@ class LineSearch(object, Loggable):
         else:
             gradient = self.fprime(self.xk + self.pk*alpha, *self.f_args)
 
-        return bare_dot(gradient, self.pk)
+        return gradient.dot(self.pk)
 
     @abc.abstractmethod
     def perform_line_search(self, xk, pk, f_k=None, fprime_k=None,

nifty/minimization/line_searching/line_search_strong_wolfe.py

@@ -136,10 +136,6 @@ class LineSearchStrongWolfe(LineSearch):
         cubic_delta = 0.2  # cubic
         quad_delta = 0.1  # quadratic
 
-        cubic_delta = 0.0  # cubic
-        quad_delta = 0.0  # quadratic
-
-
         # initialize the most recent versions (j-1) of phi and alpha
         alpha_recent = 0
         phi_recent = phi_0
@@ -170,6 +166,7 @@ class LineSearchStrongWolfe(LineSearch):
 
             # Check if the current value of alpha_j is already sufficient
             phi_alphaj = self._phi(alpha_j)
+
             # If the first Wolfe condition is not met replace alpha_hi
             # by alpha_j
             if (phi_alphaj > phi_0 + c1*alpha_j*phiprime_0) or\

nifty/minimization/steepest_descent.py

@@ -5,4 +5,9 @@ from .quasi_newton_minimizer import QuasiNewtonMinimizer
 
 class SteepestDescent(QuasiNewtonMinimizer):
     def _get_descend_direction(self, x, gradient):
-        return gradient/(-gradient.dot(gradient))
+        descend_direction = gradient
+        norm = descend_direction.norm()
+        if norm != 1:
+            return descend_direction / -norm
+        else:
+            return descend_direction * -1

nifty/minimization/vl_bfgs.py

@@ -5,8 +5,6 @@ import numpy as np
 from .quasi_newton_minimizer import QuasiNewtonMinimizer
 from .line_searching import LineSearchStrongWolfe
 
-from .bare_dot import bare_dot
-
 
 class VL_BFGS(QuasiNewtonMinimizer):
     def __init__(self, line_searcher=LineSearchStrongWolfe(), callback=None,
@@ -42,7 +40,7 @@ class VL_BFGS(QuasiNewtonMinimizer):
         for i in xrange(1, len(delta)):
             descend_direction += delta[i] * b[i]
 
-        norm = np.sqrt(bare_dot(descend_direction, descend_direction))
+        norm = descend_direction.norm()
         if norm != 1:
             descend_direction /= norm
         return descend_direction