def genetic_double_gaussian_fit(im, verbose=False) :
"fit double gaussian profiles to the each of the spectra in imcube im\nold, dont use"
shape=im.shape
X=np.arange(float(shape[0]))
res=np.empty((7,shape[1],shape[2]))
for i in xrange(shape[1]):
# f=open('log','a')
# f.write(str(i)+'\n')
# f.close()
print i
if not(verbose):
actualstdout = sys.stdout
sys.stdout = open(os.devnull,'w')
for j in xrange(shape[2]):
d=im[:,i,j]
c=sorted(d)[shape[0]/2]
peak =d.argmax()
trough=d.argmin()
mx,mn=d[peak]-c,d[trough]-c
def cons ():
r=rndint(0,2)
if r==0:
return np.array((nv(mx,mx/4), rnd()*15, nv(peak,10),\
nv(-mx,mx/4)*-10,rnd()*15, nv(peak,10),\
c*nv(1, 0.1)))
elif r==1:
return np.array((nv(mx,mx/4)*10, rnd()*15, nv(peak,10),\
nv(mn,mn/4)*-10,rnd()*15, nv(trough,10),\
c*nv(1, 0.1)))
else :
return np.array((nv(mn,mn/4)*10, rnd()*15, nv(trough,10),\
nv(mn,mn/4)*-10,rnd()*15, nv(trough,10),\
c*nv(1, 0.1)))
def foo(params):
#inlined gauss2 (this is the intermost part of the optimisation)
a1,s1,m1,a2,s2,m2,c=params
func1=np.exp(-(X-m1)**2/(2*s1*s1))
func2=np.exp(-(X-m2)**2/(2*s2*s2))
func1*=a1; func2*=a2; func1+=func2; func1+=c; func1-=d; func1*=func1
# blitz('func1=func1*a1+func2*a2+c')#not working for some reason
return func1.sum()
s=g.optimize(foo, cons, satisfactory=0.01*251, tolerance=0.001, its=100, hillWalks=2, verbose=verbose)
r=g.optimize(foo, cons, satisfactory=0.01*251, tolerance=0.001, its=100, hillWalks=2, verbose=verbose)
s.run()
r.run()
final=g.optimize(foo, cons, satisfactory=1.0, tolerance=0.0001, pool=s.pool+r.pool)
final.run()
res[:,i,j]=final.pool[0][1:]
if not(verbose):
sys.stdout.close()
sys.stdout = actualstdout
return res
def optimize(players, obj_func, initial_guess=None, verbose=False, constrained_players=None):
"""
Meta optimization using various underlying optimizers
"""
# we'll include various intermediate results in final GP for better diversity
teams_to_include = []
if initial_guess is not None:
teams_to_include.append(initial_guess)
# first try a small genetic opt to get a reasonable starting point
best_team, res = TOG.optimize(5, players, obj_func=obj_func, num_evolutions=100,
constrained_players=constrained_players, include_individuals=teams_to_include)
teams_to_include.append(copy.deepcopy(best_team))
best_obj = obj_func(best_team)
if verbose:
print("Results from initial genetic opt: {}".format(best_obj))
# now anneal that badboy
best_team = TOA.optimize(best_team, players, obj_func=obj_func, constrained_players=constrained_players)
teams_to_include.append(copy.deepcopy(best_team))
best_obj = obj_func(best_team)
if verbose:
print("Results after annealing: {}".format(best_obj))
# myopic too...
best_team = TOM.optimize(best_team, players, obj_func=obj_func, constrained_players=constrained_players)
teams_to_include.append(copy.deepcopy(best_team))
best_obj = obj_func(best_team)
# include in large scale GP
best_team, res = TOG.optimize(50, players, obj_func=obj_func, include_individuals=teams_to_include,
constrained_players=constrained_players)
best_obj = obj_func(best_team)
if verbose:
print("Results after 2nd pass genetic opt: {}".format(best_obj))
# and final myopic
best_team = TOM.optimize(best_team, players, obj_func=obj_func, constrained_players=constrained_players)
best_obj = obj_func(best_team)
if verbose:
print("Final result: {}".format(best_obj))
return best_team, best_obj
def mainfoo(im) :
shape=im.shape
X=arange(shape[0])
res=empty((7,shape[1],shape[2]))
for i in xrange(shape[1]):
f=open('log','a')
f.write(str(i)+'\n')
f.close()
print i
for j in xrange(shape[2]):
c=sorted(im[:,i,j])[shape[0]/2]
peak=im[:,i,j].argmax()
def cons ():
return array((random()*10,random()*10,random()*10-20+peak, random()*-10,random()*10,random()*20-10+peak,c*(random()*0.2+0.9)))
def foo(params):
d=im[:,i,j]
return ((gauss2(X, params)-d)**2).sum()
s=g.optimize(foo, cons, satisfactory=0.01, verbose=False)
s.run()
res[:,i,j]=s.pool[0][1:]
return res
def optimize_guest_seating(guestsCsvFile, seatsCsvFile):
"""Optimize guest seating trying to maximize sum of their preferences
Args:
guestsCsvFile (file): csv containig matrix of guest preferences
(the bigger the number the more they want to sit near)
seatsCsvFile (file): csv containing matrix representing layout of tables
Returns:
list((int, string)): list of pairs (seat_id, guest_name)
"""
guests = pd.read_table(guestsCsvFile, sep=",", index_col=0)
guest_names = guests.columns.values
seats = pd.read_table(seatsCsvFile, sep=",", index_col=0, dtype=int)
first_generation = [
np.random.permutation(range(len(guest_names))) for _ in range(10)
]
def fitness_score(chromosome):
score = 0
for i, x in enumerate(chromosome):
for j, y in enumerate(chromosome):
score += guests[guest_names[x]][guest_names[y]] * seats[str(
i)][j]
return score
seating, score = genetic.optimize(
first_generation,
genetic.selections.elite_selection,
genetic.crossovers.order_crossover,
genetic.mutations.swap_mutation,
fitness_score,
100,
0.3,
)
_logger.info("Resulting score: %s", score)
return [(i, guest_names[seating[i]]) for i in range(len(seating))]
def double_gaussian_fit_wCentral(im, verbose=False) :
"fit double gaussian profiles to the each of the spectra in imcube im"
shape=im.shape
X=np.arange(float(shape[0]))
res=np.empty((7,shape[1],shape[2]))
temp=np.empty(X.shape)
for i in xrange(shape[1]):
# f=open('log','a')
# f.write(str(i)+'\n')
# f.close()
# def blah (i):
print i
for j in xrange(shape[2]):
d=im[:,i,j]
c=sorted(d)[shape[0]/2]
temp=d.copy()
centre=shape[0]/2-8,shape[0]/2+9
temp[centre[0]:centre[1]]=[temp[centre[0]]*(1-x/16.)+temp[centre[1]]*x/16. for x in xrange (17)]
peak =temp.argmax()
trough=temp.argmin()
mx,mn=temp[peak]-c,temp[trough]-c
def cons ():
r=rndint(0,2)
if r==0:
return np.array((nv(mx,mx/4), rnd()*15, nv(peak,10),\
nv(-mx,mx/4)*-10,rnd()*15, nv(peak,10),\
c*nv(1, 0.1)))
elif r==1:
return np.array((nv(mx,mx/4)*10, rnd()*15, nv(peak,10),\
nv(mn,mn/4)*-10,rnd()*15, nv(trough,10),\
c*nv(1, 0.1)))
else :
return np.array((nv(mn,mn/4)*10, rnd()*15, nv(trough,10),\
nv(mn,mn/4)*-10,rnd()*15, nv(trough,10),\
c*nv(1, 0.1)))
def err(params,x,y):
a1,s1,m1,a2,s2,m2=params
func1=np.exp(-(x-m1)**2/(2*s1*s1))
func2=np.exp(-(x-m2)**2/(2*s2*s2))
func1*=a1; func2*=a2; func1+=func2; func1+=c; func1-=y;
return func1
v,success=optimize.leastsq(err, (d[peak]-c, 10, peak , d[trough]-d[peak], 10, trough), args=(X,temp))
if success!=1:
def foo(params):
#inlined gauss2 (this is the intermost part of the optimisation)
a1,s1,m1,a2,s2,m2,c=params
func1=np.exp(-(X-m1)**2/(2*s1*s1))
func2=np.exp(-(X-m2)**2/(2*s2*s2))
func1*=a1; func2*=a2; func1+=func2; func1+=c; func1-=d; func1*=func1
return func1.sum()
if not(verbose):
actualstdout = sys.stdout
sys.stdout = open(os.devnull,'w')
#if leastsq doesnt give a goot fit then try to fit with a genetic alg
s=g.optimize(foo, cons, satisfactory=2.51, tolerance=0.001, its=100, hillWalks=2, verbose=verbose, startWalk=True)
r=g.optimize(foo, cons, satisfactory=2.51, tolerance=0.001, its=100, hillWalks=2, verbose=verbose, startWalk=True)
s.run()
r.run()
final=g.optimize(foo, cons, satisfactory=1.0, tolerance=0.0001, pool=s.pool+r.pool, finalWalk=True)
final.run()
if not(verbose):
sys.stdout.close()
sys.stdout = actualstdout
if final.pool[0][0]<=1.5:
res[:,i,j]=final.pool[0][1:]
else:
res[:,i,j]=np.nan
else :
for k in xrange(len (v)): res[k,i,j]=v[k]
res[ -1,i,j]=c
if verbose :
plot(d)
plot (temp)
plot (gauss2(np.arange(251.), res[:,i,j]))
return res