def opt_ml(lc1, lc2, dispersionmethod, plot=False, verbose=False):
"""
microlensing optimization.
Sets the eventual microlensings of lc1 and lc2 so to minimize a dispersion.
Does not shift the curves in any other way.
If plot is True, it displays the steps of the optimization.
@type lc1: lightcurve
@param lc1: This one will be passed as first argument (i.e. "reference") to the dispersionmethod
@type lc2: lightcurve
@param lc2: The light curve that will get interpolated by the dispersionmethod.
@todo: implement a better plot ...
"""
initparams = ml.multigetfreeparams([lc1, lc2])
if verbose:
print "Initial params : ", initparams
if len(initparams) == 0:
raise RuntimeError, "There are no free microlensing params to optimize !"
if plot:
d2valuelist = []
def d2(params):
ml.multisetfreeparams([lc1, lc2], params)
ret = dispersionmethod(lc1, lc2)["d2"]
if plot:
d2valuelist.append(ret)
return ret
# http://www.scipy.org/doc/api_docs/SciPy.optimize.optimize.html#fmin_bfgs
#put disp = 0 to make them silent...
#optparams = spopt.fmin(d2, initparams)
#optparams = spopt.fmin_cg(d2, initparams)
#optparams = spopt.fmin_bfgs(d2, params, epsilon =1.0e-5)
#optparams = spopt.fmin_bfgs(d2, params, disp=1, retall=0)
#optparams = spopt.anneal(d2, params)
#optparams = spopt.fmin_ncg(d2, initparams) # this guy would need a gradient.
#optparams = spopt.fmin_bfgs(d2, initparams) # precision loss
# [ 2.89270193e-10 4.94366701e-08 -2.63101369e-04 -4.38757115e-03]
# [ 2.95294380e-10 3.64696903e-08 -2.63882092e-04 -5.66560156e-05]
optparams = spopt.fmin_powell(
d2, initparams, disp=int(verbose)
) # 312 , 2.592636, [ 2.89270193e-10 4.94366701e-08 -2.63101369e-04 -4.38757115e-03]
#optparams = spopt.fmin(d2, initparams) # 150, 2.614217
#optparams = spopt.fmin(d2, initparams, xtol = 1.0e-10)
#optparams = spopt.fmin_bfgs(d2, params, epsilon =1.0e-5)
if verbose:
print "Optimal params : ", optparams
print repr(optparams)
ml.multisetfreeparams([lc1, lc2], optparams)
if plot:
plt.semilogy(d2valuelist)
plt.show()
def dispcube(lcs, rawdispersionmethod, verbose=True, timewidth=30, timestep=1.0, optml=False, filename="dispcube.pkl"):
"""
3D specplot, calculates the dispersion over a cube of time-delays. And writes the result in a pickle.
For now this is quick and dirty programming : we want only 4 lightcurves.
This pickle can then be looked at with Mayavi (see example below)
"""
if len(lcs)!=4:
raise RuntimeError, "I want 4 lightcurves."
lcsc = [l.copy() for l in lcs]
# We apply microlensing if we do not need it.
for l in lcsc:
if optml==False and l.ml != None:
l.applyml()
# We generate the list of possible couples.
couplelist = [couple for couple in [[lc1, lc2] for lc1 in lcsc for lc2 in lcsc] if couple[0] != couple[1]]
#for couple in couplelist:
# print couple[0].object, couple[1].object
#nfree = len(lcs)-1
if optml:
pass
# def d2value(params):
# lc.multisettimedelays(lcsc, params[:nfree])
# ml.multisetfreeparams(lcsc, params[nfree:])
#
# # optimize ml here
#
# d2values = np.array([rawdispersionmethod(*couple)["d2"] for couple in couplelist])
# ret = np.mean(d2values)
# #print ret
# return ret
#
else:
def d2value(delays):
lc.multisettimedelays(lcsc, delays)
d2values = np.array([rawdispersionmethod(*couple)["d2"] for couple in couplelist])
ret = np.mean(d2values)
#print ret
return ret
initparams = np.concatenate([lc.multigettimedelays(lcsc), ml.multigetfreeparams(lcsc)])
print "Initial params : ", initparams
timeshifts = np.arange(-(timewidth)*timestep/2.0, (timewidth+1)*timestep/2.0, timestep)
cubeindexes = np.arange(timewidth + 1)
print "Points to calculate :", len(timeshifts)**3
d2cube = np.zeros((timewidth+1, timewidth+1, timewidth+1))
xshifts = timeshifts + initparams[0]
yshifts = timeshifts + initparams[1]
zshifts = timeshifts + initparams[2]
if optml==False:
for ix in cubeindexes:
print "Slice %i of %i" % (ix + 1, timewidth+1)
for iy in cubeindexes:
for iz in cubeindexes:
d2cube[ix, iy, iz] = d2value([xshifts[ix], yshifts[iy], zshifts[iz]])
# We do as if we used mgrid... no, this would be ogrid...
#xshifts = xshifts.reshape(xshifts.size, 1, 1)
#yshifts = yshifts.reshape(1, yshifts.size, 1)
#zshifts = zshifts.reshape(1, 1, zshifts.size)
beg = -(timewidth)*timestep/2.0
end = (timewidth+1)*timestep/2.0
step = timestep
x, y, z = np.mgrid[beg:end:step, beg:end:step, beg:end:step]
#print x, y, z
x += initparams[0]
y += initparams[1]
z += initparams[2]
#print x, y, z
util.writepickle({"lcs":lcs, "x":x, "y":y, "z":z, "d2":d2cube}, filename)
def dispcube(lcs,
rawdispersionmethod,
verbose=True,
timewidth=30,
timestep=1.0,
optml=False,
filename="dispcube.pkl"):
"""
3D specplot, calculates the dispersion over a cube of time-delays. And writes the result in a pickle.
For now this is quick and dirty programming : we want only 4 lightcurves.
This pickle can then be looked at with Mayavi (see example below)
"""
if len(lcs) != 4:
raise RuntimeError, "I want 4 lightcurves."
lcsc = [l.copy() for l in lcs]
# We apply microlensing if we do not need it.
for l in lcsc:
if optml == False and l.ml != None:
l.applyml()
# We generate the list of possible couples.
couplelist = [
couple for couple in [[lc1, lc2] for lc1 in lcsc for lc2 in lcsc]
if couple[0] != couple[1]
]
#for couple in couplelist:
# print couple[0].object, couple[1].object
#nfree = len(lcs)-1
if optml:
pass
# def d2value(params):
# lc.multisettimedelays(lcsc, params[:nfree])
# ml.multisetfreeparams(lcsc, params[nfree:])
#
# # optimize ml here
#
# d2values = np.array([rawdispersionmethod(*couple)["d2"] for couple in couplelist])
# ret = np.mean(d2values)
# #print ret
# return ret
#
else:
def d2value(delays):
lc.multisettimedelays(lcsc, delays)
d2values = np.array(
[rawdispersionmethod(*couple)["d2"] for couple in couplelist])
ret = np.mean(d2values)
#print ret
return ret
initparams = np.concatenate(
[lc.multigettimedelays(lcsc),
ml.multigetfreeparams(lcsc)])
print "Initial params : ", initparams
timeshifts = np.arange(-(timewidth) * timestep / 2.0,
(timewidth + 1) * timestep / 2.0, timestep)
cubeindexes = np.arange(timewidth + 1)
print "Points to calculate :", len(timeshifts)**3
d2cube = np.zeros((timewidth + 1, timewidth + 1, timewidth + 1))
xshifts = timeshifts + initparams[0]
yshifts = timeshifts + initparams[1]
zshifts = timeshifts + initparams[2]
if optml == False:
for ix in cubeindexes:
print "Slice %i of %i" % (ix + 1, timewidth + 1)
for iy in cubeindexes:
for iz in cubeindexes:
d2cube[ix, iy, iz] = d2value(
[xshifts[ix], yshifts[iy], zshifts[iz]])
# We do as if we used mgrid... no, this would be ogrid...
#xshifts = xshifts.reshape(xshifts.size, 1, 1)
#yshifts = yshifts.reshape(1, yshifts.size, 1)
#zshifts = zshifts.reshape(1, 1, zshifts.size)
beg = -(timewidth) * timestep / 2.0
end = (timewidth + 1) * timestep / 2.0
step = timestep
x, y, z = np.mgrid[beg:end:step, beg:end:step, beg:end:step]
#print x, y, z
x += initparams[0]
y += initparams[1]
z += initparams[2]
#print x, y, z
util.writepickle({
"lcs": lcs,
"x": x,
"y": y,
"z": z,
"d2": d2cube
}, filename)
def opt_ml(lc1, lc2, dispersionmethod, plot=False, verbose=False): """ microlensing optimization. Sets the eventual microlensings of lc1 and lc2 so to minimize a dispersion. Does not shift the curves in any other way. If plot is True, it displays the steps of the optimization. @type lc1: lightcurve @param lc1: This one will be passed as first argument (i.e. "reference") to the dispersionmethod @type lc2: lightcurve @param lc2: The light curve that will get interpolated by the dispersionmethod. @todo: implement a better plot ... """ initparams = ml.multigetfreeparams([lc1, lc2]) if verbose: print "Initial params : ", initparams if len(initparams) == 0: raise RuntimeError, "There are no free microlensing params to optimize !" if plot: d2valuelist = [] def d2(params): ml.multisetfreeparams([lc1, lc2], params) ret = dispersionmethod(lc1, lc2)["d2"] if plot: d2valuelist.append(ret) return ret # http://www.scipy.org/doc/api_docs/SciPy.optimize.optimize.html#fmin_bfgs #put disp = 0 to make them silent... #optparams = spopt.fmin(d2, initparams) #optparams = spopt.fmin_cg(d2, initparams) #optparams = spopt.fmin_bfgs(d2, params, epsilon =1.0e-5) #optparams = spopt.fmin_bfgs(d2, params, disp=1, retall=0) #optparams = spopt.anneal(d2, params) #optparams = spopt.fmin_ncg(d2, initparams) # this guy would need a gradient. #optparams = spopt.fmin_bfgs(d2, initparams) # precision loss # [ 2.89270193e-10 4.94366701e-08 -2.63101369e-04 -4.38757115e-03] # [ 2.95294380e-10 3.64696903e-08 -2.63882092e-04 -5.66560156e-05] optparams = spopt.fmin_powell(d2, initparams, disp=int(verbose)) # 312 , 2.592636, [ 2.89270193e-10 4.94366701e-08 -2.63101369e-04 -4.38757115e-03] #optparams = spopt.fmin(d2, initparams) # 150, 2.614217 #optparams = spopt.fmin(d2, initparams, xtol = 1.0e-10) #optparams = spopt.fmin_bfgs(d2, params, epsilon =1.0e-5) if verbose: print "Optimal params : ", optparams print repr(optparams) ml.multisetfreeparams([lc1, lc2], optparams) if plot: plt.semilogy(d2valuelist) plt.show()
