def gen_samples(wdir,nsamples=100,nsize=10):
load_config('%s/input.py'%wdir)
conf['bootstrap']=False
istep=core.get_istep()
core.mod_conf(istep) #--set conf as specified in istep
resman=RESMAN(nworkers=5,parallel=True,datasets=True)
parman=resman.parman
order=parman.order
obsres={}
if 'idis' in conf['datasets'] : obsres['idis'] = resman.idisres
if 'pidis' in conf['datasets'] : obsres['pidis'] = resman.pidisres
if 'sidis' in conf['datasets'] : obsres['sidis'] = resman.sidisres
if 'psidis' in conf['datasets'] : obsres['psidis'] = resman.psidisres
if 'dy' in conf['datasets'] : obsres['dy'] = resman.dyres
if 'sia' in conf['datasets'] : obsres['sia'] = resman.siares
print('\ngen ml samples using the setup as in %s\n'%wdir)
checkdir('%s/mlsamples'%wdir)
data=gen_dataframe(parman.order,obsres,nsize)
for _ in range(nsamples):
lprint('progress: %d/%d'%(_,nsamples))
par=parman.gen_flat(setup=True)
res,rres,nres=resman.get_residuals(par)
#--fill dataframe
i=data['size']
data['params'][i]=par
for reaction in obsres:
for idx in data['reactions'][reaction]:
prediction=copy.copy(obsres[reaction].tabs[idx]['prediction'])
data['reactions'][reaction][idx][i]=prediction
data['size']+=1
if data['size']==nsize:
save(data,'%s/mlsamples/%s'%(wdir,id_generator(12)))
data=gen_dataframe(parman.order,obsres,nsize)
print
#--close resman
resman.shutdown()
def get_msr_inspected(wdir):
print('\nget msr inspected (filtered msr files) using %s\n'%wdir)
replicas=sorted(os.listdir('%s/msr'%wdir))
X=[]
for i in range(len(replicas)):
lprint('progress: %d/%d'%(i+1,len(replicas)))
replica=load('%s/msr/%s'%(wdir,replicas[i]))
istep=sorted(replica['params'].keys())[-1] #--pick last step
params=replica['params'][istep]
#print params
#if params[2]<0.1: continue
if params is None: continue
data=replica['chi2']#[istep]
chi2,npts=0,0
for reaction in data:
#print
for idx in data[reaction]:
chi2+=data[reaction][idx]['chi2']
npts+=data[reaction][idx]['npts']
if chi2/npts-1<1000:
X.append(chi2/npts-1)
checkdir('%s/msr-inspected'%wdir)
cmd=['cp']
cmd.append('%s/msr/%s'%(wdir,replicas[i]))
cmd.append('%s/msr-inspected/%s'%(wdir,replicas[i]))
p = subprocess.Popen(cmd, stdout=subprocess.PIPE)
print()
print('original num. samples :%d'%len(replicas))
print('inspected num. samples :%d'%len(X))
nrows=1
ncols=1
#--plot labeled residuals
ax=py.subplot(nrows,ncols,1)
ax.hist(X,bins=10)
#py.tight_layout()
checkdir('%s/gallery'%wdir)
py.savefig('%s/gallery/chi2-dist-inspect.pdf'%(wdir))
py.close()
def run(self):
resman = conf['resman']
parman = conf['parman']
args = conf['args']
a = timer()
np.random.seed(12345)
for i in range(10):
msg = '[%d/10]' % i
lprint(msg)
R = np.random.randn(len(parman.par))
par = parman.par + 1e-3 * R
res, rres, nres = resman.get_residuals(par)
print '\nt-elapsed (sec): ', (timer() - a) / 10
print 'npts=%d' % res.size
print 'chi2=%f' % np.sum(res**2)
print 'chi2/npts=%f' % (np.sum(res**2) / res.size)
print 'chi2(r)=%f' % np.sum(rres**2)
print 'chi2(norm)=%f' % np.sum(nres**2)
def rename_tables(dirname, newname):
print('\nrenaming LHAPDF tables for %s to %s' % (dirname, newname))
checkdir(newname)
F = os.listdir(dirname)
cnt = 0
for f in F:
cnt += 1
lprint('progress %d/%d' % (cnt, len(F)))
cmd = [
'cp',
'%s/%s' % (dirname, f),
'%s/%s' % (newname, f.replace(dirname, newname))
]
p = Popen(cmd, stdout=PIPE, stderr=STDOUT)
output = p.stdout.read()
print
def gen_dist_from_fitpack(wdir,dist):
print('\ngenerating %s tables for benchmark using %s'%(dist,wdir))
load_config('%s/input.py'%wdir)
istep=core.get_istep()
core.mod_conf(istep) #--set conf as specified in istep
resman=RESMAN(nworkers=1,parallel=False,datasets=False)
parman=resman.parman
jar=load('%s/data/jar-%d.dat'%(wdir,istep))
replicas=jar['replicas']
X,Q2=gen_grid(dist)
qpd=conf[dist]
nx=len(X)
nQ2=len(Q2)
flavs=[-5,-4,-3,-2,-1,1,2,3,4,5,21]
fmap={-5:'bb',-4:'cb',-3:'sb',-2:'ub',-1:'db'}
fmap.update({5:'b',4:'c',3:'s',2:'u',1:'d'})
fmap.update({21:'g'})
#--gen qpd data per replica
checkdir('%s/benchmark'%wdir)
cnt=0
for par in replicas:
lprint('progress: %d/%d'%(cnt+1,len(replicas)))
parman.set_new_params(par)
data={}
for _ in Q2:
data[_]={}
for flav in flavs:
data[_][flav]=np.array([qpd.get_xF(x,_,fmap[flav]) for x in X])
save(data,'%s/benchmark/%s-%d.dat'%(wdir,dist,cnt))
cnt+=1
print
def gen_priors(wdir, kc, nsamples):
load_config('%s/input.py' % wdir)
istep = core.get_istep()
#cluster,colors,nc,cluster_order = core.get_clusters(wdir,istep)
cluster, colors, nc, cluster_order = classifier.get_clusters(
wdir, istep, kc)
best_cluster = cluster_order[0]
#--get data
replicas = core.get_replicas(wdir)
step = conf['steps'][istep]
samples = []
for i in range(len(replicas)):
if cluster[i] != best_cluster: continue
replica = replicas[i]
order = replica['order'][istep]
params = replica['params'][istep]
samples.append(params)
samples = np.array(samples)
pmin = [np.amin(p) for p in samples.T]
pmax = [np.amax(p) for p in samples.T]
new = []
for i in range(len(pmin)):
new.append(np.random.uniform(low=pmin[i], high=pmax[i], size=nsamples))
new = np.transpose(new)
checkdir('%s/msr-opt-priors' % wdir)
replica = replicas[0] #--template replica
for i in range(nsamples):
lprint('%d/%d' % (i + 1, nsamples))
replica['params'][istep] = new[i]
fname = '%s/msr-opt-priors/%s.msr' % (wdir, id_generator(12))
save(replica, fname)
print
def gen_dist_from_lhapdf(wdir,dist,dirname):
print('\ngenerating %s tables for benchmark using %s from lhapdf'%(dist,wdir))
replicas=lhapdf.mkPDFs(dirname)
X,Q2=gen_grid(dist)
flavs=[-5,-4,-3,-2,-1,1,2,3,4,5,21]
#--gen qpd data per replica
checkdir('%s/benchmark'%wdir)
cnt=0
for i in range(len(replicas)):
lprint('progress: %d/%d'%(cnt+1,len(replicas)))
replica=replicas[i]
data={}
for _ in Q2:
data[_]={}
for flav in flavs:
data[_][flav]=np.array([replica.xfxQ2(flav,x,_) for x in X])
save(data,'%s/benchmark/%s-%d-lha.dat'%(wdir,dist,cnt))
cnt+=1
print
def gen_tables(wdir, dist, file_name, info, info_only=False):
print('\ngenerating LHAPDF tables for %s using %s' % (dist, wdir))
load_config('%s/input.py' % wdir)
istep = core.get_istep()
core.mod_conf(istep) #--set conf as specified in istep
resman = RESMAN(nworkers=1, parallel=False, datasets=False)
parman = resman.parman
jar = load('%s/data/jar-%d.dat' % (wdir, istep))
replicas = jar['replicas']
#--check order consistency
order = jar['order']
parman.order = order
#--create output dir
checkdir(wdir + '/data/')
checkdir(wdir + '/data/%s/' % file_name)
#--gen lhapdf_data_files
if info_only == False:
cnt = 0
for par in replicas:
lprint('progress: %d/%d' % (cnt + 1, len(replicas)))
parman.set_new_params(par)
X, Q2, table = _gen_table(dist)
gen_lhapdf_dat_file(X, Q2, table, wdir, file_name, cnt)
cnt += 1
print
#--gen_lhapdf_info_file
X, Q2 = gen_grid(dist)
nrep = len(replicas)
gen_lhapdf_info_file(X, Q2, nrep, wdir, file_name, info)
def get_predictions(wdir, cores=2, mod_conf=None, pre_flags={}):
## 'pre_flags' can be defined for example like the following
## pre_flags = {'regenerate_same': False, 'regenerate_part': False, 'regenerate_mismatch': True}
if mod_conf == None:
load_config('%s/input.py' % wdir)
else:
config.conf = conf
conf['bootstrap'] = False
istep = core.get_istep()
current_replica_names = sorted(os.listdir('%s/msr-inspected' % wdir))
previous_replica_names_path = '%s/data/predicted-replicas-%d.dat' % (wdir,
istep)
previous_predictions_path = '%s/data/predictions-%d.dat' % (wdir, istep)
part_flag = False
if os.path.exists(previous_predictions_path) and os.path.exists(
previous_replica_names_path):
previous_replica_names = load(previous_replica_names_path)
if current_replica_names == previous_replica_names:
print(
'The replicas names are the same since predictions were generated last time,'
)
print(
'do you wish to regenrate and overwrite previous predictions?')
if 'regenerate_same' in pre_flags:
if pre_flags['regenerate_same']:
print(
'regenerating everything based on predifined flag...')
part_flag = False
else:
print(
'exit generating predictions based on predifined flag.'
)
return
else:
if sys.version_info[0] == 2:
while True:
same_flag = raw_input(
'y (regenerate)/n (skip generation of predictions): '
)
if same_flag == '':
print('please type "y" or "n"')
elif same_flag.lower()[0] == 'n':
return
elif same_flag.lower()[0] == 'y':
part_flag = False
break
else:
print('please type "y" or "n"')
elif sys.version_info[0] == 3:
while True:
same_flag = input(
'y (regenerate)/n (skip generation of predictions): '
)
if same_flag == '':
print('please type "y" or "n"')
elif same_flag.lower()[0] == 'n':
return
elif same_flag.lower()[0] == 'y':
part_flag = False
break
else:
print('please type "y" or "n"')
elif all(_ in current_replica_names for _ in previous_replica_names):
print('Based on the replica names,')
print(
'part of the predictions you are trying to generate has already been generated last time,'
)
print(
'do you wish to regenrate with all replicas or only generate with new replicas?'
)
if 'regenerate_part' in pre_flags:
if pre_flags['regenerate_part']:
print(
'regenerating with all replicas based on predifined flag...'
)
part_flag = False
else:
print(
'generating only on new replicas based on predifined flag...'
)
part_flag = True
else:
if sys.version_info[0] == 2:
while True:
part_flag = raw_input(
'w (regenerate with all replicas)/p (generate only with new replicas): '
)
if part_flag == '':
print('please type "w" or "p"')
elif part_flag.lower()[0] == 'w':
part_flag = False
break
elif part_flag.lower()[0] == 'p':
part_flag = True
break
else:
print('please type "y" or "n"')
elif sys.version_info[0] == 3:
while True:
part_flag = input(
'w (regenerate with all replicas)/p (generate only with new replicas): '
)
if part_flag == '':
print('please type "w" or "p"')
elif part_flag.lower()[0] == 'w':
part_flag = False
break
elif part_flag.lower()[0] == 'p':
part_flag = True
break
else:
print('please type "y" or "n"')
else:
part_flag = False
else:
pass
if part_flag:
replicas_to_load = np.setdiff1d(current_replica_names,
previous_replica_names)
replicas = [
load('%s/msr-inspected/%s' % (wdir, _)) for _ in replicas_to_load
]
else:
replicas = core.get_replicas(wdir)
core.mod_conf(istep, replicas[0]) #--set conf as specified in istep
resman = RESMAN(nworkers=cores, parallel=True, datasets=True)
parman = resman.parman
order = replicas[0]['order'][istep]
## order and parameters are updated later for each replicas individually
# parman.order = order
# order = parman.order
if part_flag:
previous_data = load(previous_predictions_path)
if order != previous_data['order']:
print(
'Previous replicas have different parameter orders from new replicas,'
)
print('previously: ', previous_data['order'])
print('currently: ', order)
print(
'do you wish to overwrite the previous predictions or exit to check?'
)
if 'regenerate_mismatch' in pre_flags:
if pre_flags['regenerate_mismatch']:
print(
'regenerating everything based on predifined flag...')
part_flag = False
exit_flag = False
else:
print(
'exit generating predictions based on predifined flag')
exit_flag = True
else:
part_flag, exit_flag = ask_replica_mismatch()
if exit_flag: sys.exit()
obsres = {}
if 'idis' in conf['datasets']: obsres['idis'] = resman.idis_res
if 'pidis' in conf['datasets']: obsres['pidis'] = resman.pidis_res
if 'sidis' in conf['datasets']: obsres['sidis'] = resman.sidis_res
if 'psidis' in conf['datasets']: obsres['psidis'] = resman.psidis_res
if 'dy' in conf['datasets']: obsres['dy'] = resman.dy_res
if 'sia' in conf['datasets']: obsres['sia'] = resman.sia_res
if 'qpdf' in conf['datasets']: obsres['qpdf'] = resman.qpdf_res
if 'dy-pion' in conf['datasets']: obsres['dy-pion'] = resman.dy_pion_res
if 'pion_qT' in conf['datasets']: obsres['pion_qT'] = resman.pion_qTres
if 'ln' in conf['datasets']: obsres['ln'] = resman.ln_res
if 'jet' in conf['datasets']: obsres['jet'] = resman.jet_res
if 'pjet' in conf['datasets']: obsres['pjet'] = resman.pjet_res
if part_flag:
# previous_data = load(previous_predictions_path)
if obsres.keys() != previous_data['reactions'].keys():
print(
'Previous replicas have different observables from new replicas,'
)
print('previously: ', previous_data['reactions'].keys())
print('currently: ', obsres.keys())
print(
'do you wish to overwrite the previous predictions or exit to check?'
)
if 'regenerate_mismatch' in pre_flags:
if pre_flags['regenerate_mismatch']:
print(
'regenerating everything based on predifined flag...')
part_flag = False
exit_flag = False
else:
print(
'exit generating predictions based on predifined flag')
exit_flag = True
else:
part_flag, exit_flag = ask_replica_mismatch()
if exit_flag: sys.exit()
if part_flag:
for _ in obsres.keys():
if obsres[_].tabs.keys() != previous_data['reactions'][_].keys(
):
print(
'Previous replicas have different datasets from new replicas in observable %s,'
% _)
print('previously: ', previous_data['reactions'][_].keys())
print('currently: ', obsres[_].tabs.keys())
print(
'do you wish to overwrite the previous predictions or exit to check?'
)
if 'regenerate_mismatch' in pre_flags:
if pre_flags['regenerate_mismatch']:
print(
'regenerating everything based on predifined flag...'
)
part_flag = False
exit_flag = False
else:
print(
'exit generating predictions based on predifined flag'
)
exit_flag = True
else:
part_flag, exit_flag = ask_replica_mismatch()
if exit_flag: sys.exit()
#--setup big table to store all we want
data = {}
data['order'] = order
data['params'] = []
data['reactions'] = {}
data['res'] = []
data['rres'] = []
data['nres'] = []
for _ in obsres:
tabs = copy.copy(obsres[_].tabs)
#--create a space to store all the predictions from replicas
for idx in tabs:
tabs[idx]['prediction-rep'] = []
tabs[idx]['residuals-rep'] = []
data['reactions'][_] = tabs
print('\ngen predictions using %s\n' % wdir)
cnt = 0
if part_flag:
n_all_replicas = len(current_replica_names)
n_available_replicas = len(previous_replica_names)
cnt += n_available_replicas
else:
replicas = core.get_replicas(wdir)
for replica in replicas:
cnt += 1
if part_flag:
lprint('progress: %d/%d' % (cnt, n_all_replicas))
else:
lprint('progress: %d/%d' % (cnt, len(replicas)))
core.mod_conf(istep, replica) #--will update passive dist
parman.par = copy.copy(replica['params'][istep])
parman.order = copy.copy(replica['order'][istep])
data['params'] = np.append(data['params'], parman.par)
#for i in range(len(parman.par)):
# print parman.order[i],parman.par[i],replica['order'][istep][i]
#--compute residuals (==theory)
res, rres, nres = resman.get_residuals(parman.par, initial=True)
data['res'].append(res)
data['rres'].append(rres)
data['nres'].append(nres)
#--save predictions of the current step and current replica at data
for _ in obsres:
for idx in data['reactions'][_]:
prediction = copy.copy(obsres[_].tabs[idx]['prediction'])
residuals = copy.copy(obsres[_].tabs[idx]['residuals'])
data['reactions'][_][idx]['prediction-rep'].append(prediction)
data['reactions'][_][idx]['residuals-rep'].append(residuals)
print
#--close resman
resman.shutdown()
## append newly generated parts if 'part_flag' is True
if part_flag:
data['params'] = np.append(previous_data['params'], data['params'])
data['res'] = np.concatenate((previous_data['res'], data['res']))
data['rres'] = np.concatenate((previous_data['rres'], data['rres']))
data['nres'] = np.concatenate((previous_data['nres'], data['nres']))
for _ in data['reactions']:
for dataset in data['reactions'][_]:
data['reactions'][_][dataset][
'prediction-rep'] = previous_data['reactions'][_][dataset][
'prediction-rep'] + data['reactions'][_][dataset][
'prediction-rep']
data['reactions'][_][dataset]['residuals-rep'] = previous_data[
'reactions'][_][dataset]['prediction-rep'] + data[
'reactions'][_][dataset]['residuals-rep']
#--convert tables to numpy array before saving
for _ in ['res', 'rres', 'nres']:
data[_] = np.array(data[_])
checkdir('%s/data' % wdir)
if mod_conf == None:
save(data, '%s/data/predictions-%d.dat' % (wdir, istep))
save(current_replica_names,
'%s/data/predicted-replicas-%d.dat' % (wdir, istep))
else:
save(data, '%s/data/predictions-%d-sim.dat' % (wdir, istep))
def get_moments(wdir, flavors, x_1, x_2, Q2=None):
load_config('%s/input.py' % wdir)
istep = core.get_istep()
replicas = core.get_replicas(wdir)
## 'conf' will be modified for each replica individually later in the loop over 'replicas'
## the reason for doing this is that 'fix parameters' has to be set correctly for each replica
if 'ppdf_smx' not in conf['steps'][istep]['active distributions']:
print('ppdf-smx-proton not in active distribution')
return
resman = RESMAN(nworkers=1, parallel=False, datasets=False)
parman = resman.parman
parman.order = replicas[0]['order'][istep]
## make sure 'parman' uses the same order for active distributions as all the replicas do
# print parman.order
ppdf = conf['ppdf_smx']
## setup kinematics
# X = 10.0 ** np.linspace(-3, -1, 100)
# X = np.append(X, np.linspace(0.1, 0.99, 100))
if Q2 == None: Q2 = conf['Q20']
print(
'\ngenerating momentus for polarized pdf_smx-proton from %s at Q2 = %f'
% (wdir, Q2))
## compute moments for all replicas
moments = {}
statistics = {}
n_replicas = len(replicas)
for i in range(n_replicas):
lprint('%d/%d' % (i + 1, n_replicas))
parman.order = copy.copy(replicas[i]['order'][istep])
core.mod_conf(istep, replicas[i])
parman.set_new_params(replicas[i]['params'][istep], initial=True)
for flavor in flavors:
if flavor not in moments: moments[flavor] = []
if flavor not in statistics: statistics[flavor] = {}
if flavor == 'up':
func = lambda x: (ppdf.get_xF(x, Q2, 'u') + ppdf.get_xF(
x, Q2, 'ub')) / x
elif flavor == 'dp':
func = lambda x: (ppdf.get_xF(x, Q2, 'd') + ppdf.get_xF(
x, Q2, 'sb')) / x
elif flavor == 'sp':
func = lambda x: (ppdf.get_xF(x, Q2, 's') + ppdf.get_xF(
x, Q2, 'sb')) / x
else:
func = lambda x: ppdf.get_xF(x, Q2, flavor) / x
moments[flavor].append(quad(func, x_1, x_2)[0])
for flavor in flavors:
statistics[flavor]['mean'] = np.mean(moments[flavor])
statistics[flavor]['standard_deviation'] = np.std(moments[flavor])
print
for flavor in flavors:
print flavor, ': ', statistics[flavor]
checkdir('%s/data' % wdir)
if Q2 == conf['Q20']:
save({
'Q2': Q2,
'moments': moments,
'statistics': statistics
}, '%s/data/ppdf_smx-moments-%s-to-%s-%d.dat' %
(wdir, x_1, x_2, istep))
else:
save({
'Q2': Q2,
'moments': moments,
'statistics': statistics
}, '%s/data/ppdf_smx-moments-%s-to-%s-%d-%f.dat' %
(wdir, x_1, x_2, istep, Q2))
def gen_xf(wdir, flavors, Q2=None):
load_config('%s/input.py' % wdir)
istep = core.get_istep()
replicas = core.get_replicas(wdir)
## 'conf' will be modified for each replica individually later in the loop over 'replicas'
## the reason for doing this is that 'fix parameters' has to be set correctly for each replica
if 'ppdf_smx' not in conf['steps'][istep]['active distributions']:
print('ppdf_smx-proton not in active distribution')
return
resman = RESMAN(nworkers=1, parallel=False, datasets=False)
parman = resman.parman
parman.order = replicas[0]['order'][istep]
## make sure 'parman' uses the same order for active distributions as all the replicas do
# print parman.order
ppdf = conf['ppdf_smx']
ppdfJAM = conf['ppdf']
## setup kinematics
xs = 10.0**np.linspace(-5, -1, 100)
xs = np.append(xs, np.linspace(0.1, 0.3, 100))
#xs = np.append(xs, np.linspace(0.1, 0.99, 100))
if Q2 == None: Q2 = conf['Q20']
print('\ngenerating polarized pdf-proton from %s at Q2 = %f' % (wdir, Q2))
## compute xf for all replicas
xfs = {}
n_replicas = len(replicas)
for i in range(n_replicas):
lprint('%d/%d' % (i + 1, n_replicas))
## filter
#flag=False
#params=replica['params'][istep]
#order=replica['order'][istep]
#for i in range(len(order)):
# if order[i][0]!=1:continue
# if order[i][1]!='ppdf':continue
# #if order[i][2]=='s1 a':
# # if params[i]<-0.9: flag=True
#if flag: continue
parman.order = copy.copy(replicas[i]['order'][istep])
core.mod_conf(istep, replicas[i])
parman.set_new_params(replicas[i]['params'][istep], initial=True)
x0 = conf['smx']['x0']
xsep = conf['smx']['xsep']
for flavor in flavors:
if flavor not in xfs: xfs[flavor] = []
if flavor == 'up':
#func = lambda x: x*ppdf.get_C(x, Q2)[1]
#func = lambda x: np.heaviside(xsep-x,1)*(x*ppdf.get_C(x, Q2)[1] + ppdfJAM.get_xF(xsep,xsep*Q2/x,flavor)) + np.heaviside(x-xsep,0)*ppdfJAM.get_xF(x,Q2,flavor)
func = lambda x: np.heaviside(xsep - x, 1) * (x * ppdf.get_C(
x, Q2)[1] + x * ppdf.get_C_NOevo(xsep, xsep * Q2 / x)[
1]) + np.heaviside(x - xsep, 0) * x * ppdf.get_C_NOevo(
x, Q2)[1]
elif flavor == 'dp':
#func = lambda x: x*ppdf.get_C(x, Q2)[2]
#func = lambda x: np.heaviside(xsep-x,1)*(x*ppdf.get_C(x, Q2)[2] + ppdfJAM.get_xF(xsep,xsep*Q2/x,flavor)) + np.heaviside(x-xsep,0)*ppdfJAM.get_xF(x,Q2,flavor)
func = lambda x: np.heaviside(xsep - x, 1) * (x * ppdf.get_C(
x, Q2)[2] + x * ppdf.get_C_NOevo(xsep, xsep * Q2 / x)[
2]) + np.heaviside(x - xsep, 0) * x * ppdf.get_C_NOevo(
x, Q2)[2]
elif flavor == 'sp':
#func = lambda x: x*ppdf.get_C(x, Q2)[3]
#func = lambda x: np.heaviside(xsep-x,1)*(x*ppdf.get_C(x, Q2)[3] + ppdfJAM.get_xF(xsep,xsep*Q2/x,flavor)) + np.heaviside(x-xsep,0)*ppdfJAM.get_xF(x,Q2,flavor)
func = lambda x: np.heaviside(xsep - x, 1) * (x * ppdf.get_C(
x, Q2)[3] + x * ppdf.get_C_NOevo(xsep, xsep * Q2 / x)[
3]) + np.heaviside(x - xsep, 0) * x * ppdf.get_C_NOevo(
x, Q2)[3]
elif flavor == 'g':
#func = lambda x: x*ppdf.get_C(x, Q2)[0]
#func = lambda x: np.heaviside(xsep-x,1)*(x*ppdf.get_C(x, Q2)[0] + ppdfJAM.get_xF(xsep,xsep*Q2/x,flavor)) + np.heaviside(x-xsep,0)*ppdfJAM.get_xF(x,Q2,flavor)
func = lambda x: np.heaviside(xsep - x, 1) * (x * ppdf.get_C(
x, Q2)[0] + x * ppdf.get_C_NOevo(xsep, xsep * Q2 / x)[
0]) + np.heaviside(x - xsep, 0) * x * ppdf.get_C_NOevo(
x, Q2)[0]
xfs[flavor].append([func(x) for x in xs])
print
checkdir('%s/data' % wdir)
if Q2 == conf['Q20']:
save({
'X': xs,
'Q2': Q2,
'XF': xfs
}, '%s/data/ppdf_smx-%d.dat' % (wdir, istep))
else:
save({
'X': xs,
'Q2': Q2,
'XF': xfs
}, '%s/data/ppdf_smx-%d-%f.dat' % (wdir, istep, Q2))
def get_parameters(wdir, Q2=None):
load_config('%s/input.py' % wdir)
istep = core.get_istep()
replicas = core.get_replicas(wdir)
## 'conf' will be modified for each replica individually later in the loop over 'replicas'
## the reason for doing this is that 'fix parameters' has to be set correctly for each replica
if 'ppdf_smx' not in conf['steps'][istep]['active distributions']:
print('ppdf_smx-proton not in active distribution')
return
resman = RESMAN(nworkers=1, parallel=False, datasets=False)
parman = resman.parman
parman.order = replicas[0]['order'][
istep] ## make sure 'parman' uses the same order as all the replicas do
# print parman.order
ppdf = conf['ppdf_smx']
## setup kinematics
if Q2 == None: Q2 = conf['Q20']
## get parameters for all flavors of PDF
print('\ngetting ppdf_smx-parameters from %s at Q2 = %.2f' % (wdir, Q2))
## check if any of the shapes are fixed
shape_1 = []
shape_2 = []
shape_3 = []
for parameter in conf['params']['ppdf_smx']:
if '1' in parameter:
shape_1.append(conf['params']['ppdf_smx'][parameter]['fixed'])
elif '2' in parameter:
shape_2.append(conf['params']['ppdf_smx'][parameter]['fixed'])
elif '3' in parameter:
shape_3.append(conf['params']['ppdf_smx'][parameter]['fixed'])
else:
print('there seems to be more than three shapes')
print(parameter)
sys.exit('please update script %s' % __file__)
fixed_shape_1 = False
fixed_shape_2 = False
fixed_shape_3 = False
if (len(shape_1) != 0) and all([_ == True for _ in shape_1]):
fixed_shape_1 = True
elif len(shape_1) == 0:
fixed_shape_1 = True
if (len(shape_2) != 0) and all([_ == True for _ in shape_2]):
fixed_shape_2 = True
elif len(shape_2) == 0:
fixed_shape_2 = True
if (len(shape_3) != 0) and all([_ == True for _ in shape_3]):
fixed_shape_3 = True
elif len(shape_3) == 0:
fixed_shape_3 = True
parameters = {}
if not fixed_shape_1: parameters[1] = {}
if not fixed_shape_2: parameters[2] = {}
if not fixed_shape_3: parameters[3] = {}
n_replicas = len(replicas)
for i in range(n_replicas):
lprint('%d/%d' % (i + 1, n_replicas))
parman.order = copy.copy(replicas[i]['order'][istep])
core.mod_conf(istep, replicas[i])
parman.set_new_params(replicas[i]['params'][istep], initial=True)
for flavor, value in ppdf.params.iteritems():
for j in parameters:
if str(j) in flavor:
flavor_key = flavor.replace(str(j), '')
if flavor_key not in parameters[j]:
parameters[j][flavor_key] = {
'n': [],
'a': [],
'b': [],
'c': [],
'd': []
}
parameters[j][flavor_key]['n'].append(value[0])
parameters[j][flavor_key]['a'].append(value[1])
parameters[j][flavor_key]['b'].append(value[2])
parameters[j][flavor_key]['c'].append(value[3])
parameters[j][flavor_key]['d'].append(value[4])
else:
pass
## remove c or d parameters if fixed in all flavors and shapes
## this is to avoid producing empty figures
shapes_fixed_c = []
shapes_fixed_d = []
for shape in parameters:
flavors_fixed_c = []
flavors_fixed_d = []
for flavor in parameters[shape]:
cs = parameters[shape][flavor]['c']
ds = parameters[shape][flavor]['d']
if all([_ == cs[0] for _ in cs]):
flavors_fixed_c.append(True)
else:
flavors_fixed_c.append(False)
if all([_ == ds[0] for _ in ds]):
flavors_fixed_d.append(True)
else:
flavors_fixed_d.append(False)
if all(flavors_fixed_c):
shapes_fixed_c.append(True)
else:
shapes_fixed_c.append(False)
if all(flavors_fixed_d):
shapes_fixed_d.append(True)
else:
shapes_fixed_d.append(False)
if all(shapes_fixed_c):
for shape in parameters:
for flavor in parameters[shape]:
del parameters[shape][flavor]['c']
print('parameter c is fixed for shape %s' % shape)
if all(shapes_fixed_d):
for shape in parameters:
for flavor in parameters[shape]:
del parameters[shape][flavor]['d']
print('parameter d is fixed for shape %s' % shape)
print
checkdir('%s/data' % wdir)
if Q2 == conf['Q20']:
save(parameters, '%s/data/ppdf_smx-parameters-%d.dat' % (wdir, istep))
else:
save(parameters,
'%s/data/ppdf_smx-parameters-%d-%f.dat' % (wdir, istep, Q2))
def gen_xf(wdir, had, flavors=['g', 'u', 'ub', 'd', 'db', 's', 'sb'], Q2=None):
fflabel = 'ff%s' % had
print('\ngenerating ff-%s from %s' % (had, wdir))
load_config('%s/input.py' % wdir)
istep = core.get_istep()
replicas = core.get_replicas(wdir)
core.mod_conf(istep, replicas[0]) ## set conf as specified in istep
if fflabel not in conf['steps'][istep]['active distributions']:
print('ff-%s not in active distribution' % had)
return
resman = RESMAN(nworkers=1, parallel=False, datasets=False)
parman = resman.parman
jar = load('%s/data/jar-%d.dat' % (wdir, istep))
## 'jar' contains parameters and their orders
replicas = jar['replicas']
## 'jar['replicas']' is a list, in which each element is a list of parameters obeying 'jar['order']'
ff = conf[fflabel]
## setup kinematics
X = np.linspace(0.01, 0.99, 100)
if Q2 == None: Q2 = conf['Q20']
## compute XF for all replicas
XF = {}
cnt = 0
for par in replicas:
cnt += 1
lprint('%d/%d' % (cnt, len(replicas)))
## filter
#flag=False
#params=replica['params'][istep]
#order=replica['order'][istep]
#for i in range(len(order)):
# if order[i][0]!=1:continue
# if order[i][1]!='pdf':continue
# #if order[i][2]=='s1 a':
# # if params[i]<-0.9: flag=True
#if flag: continue
parman.set_new_params(par, initial=True)
#print
#print conf['ffpion'].get_xF(0.5,10.0,'u')
#print conf['ffkaon'].get_xF(0.5,10.0,'u')
#print ff.get_xF(0.5,10.0,'u')
for flavor in flavors:
if flavor not in XF: XF[flavor] = []
if flavor == 'c' or flavor == 'cb' or flavor == 'c+cb':
_Q2 = conf['aux'].mc2 + 1
elif flavor == 'b' or flavor == 'bb' or flavor == 'b+bb':
_Q2 = conf['aux'].mb2 + 1
else:
_Q2 = Q2
if flavor == 'u+ub':
func = lambda x: ff.get_xF(x, _Q2, 'u') + ff.get_xF(
x, _Q2, 'ub')
elif flavor == 'd+db':
func = lambda x: ff.get_xF(x, _Q2, 'd') + ff.get_xF(
x, _Q2, 'db')
elif flavor == 's+sb':
func = lambda x: ff.get_xF(x, _Q2, 's') + ff.get_xF(
x, _Q2, 'cb')
elif flavor == 'c+cb':
func = lambda x: ff.get_xF(x, _Q2, 'c') + ff.get_xF(
x, _Q2, 'cb')
elif flavor == 'b+bb':
func = lambda x: ff.get_xF(x, _Q2, 'b') + ff.get_xF(
x, _Q2, 'bb')
else:
func = lambda x: ff.get_xF(x, _Q2, flavor)
XF[flavor].append([func(x) for x in X])
#print func(0.5)
print
checkdir('%s/data' % wdir)
if Q2 == conf['Q20']:
save({
'X': X,
'Q2': Q2,
'XF': XF
}, '%s/data/ff%s-%d.dat' % (wdir, had, istep))
else:
save({
'X': X,
'Q2': Q2,
'XF': XF
}, '%s/data/ff%s-%d-%d.dat' % (wdir, had, istep, int(Q2)))
def compare2(wdir,dist,title,xscale='log'):
F=os.listdir('%s/benchmark/'%(wdir))
true = [_ for _ in F if '-lha' not in _ if dist in _]
lha = [_ for _ in F if '-lha' in _ if dist in _]
itrue=np.argsort([int(_.split('-')[-1].split('.')[0]) for _ in true])
true=[true[i] for i in itrue]
ilha=np.argsort([int(_.replace('-lha','').split('-')[-1].split('.')[0]) for _ in lha])
lha=[lha[i] for i in ilha]
X,Q2=gen_grid(dist)
flavs=[21,2,1,3,-2,-1,-3,4,5]
fmap={}
fmap[21] =r'$g$'
fmap[ 2] =r'$u$'
fmap[ 1] =r'$d$'
fmap[ 3] =r'$s$'
fmap[ 4] =r'$c$'
fmap[ 5] =r'$b$'
fmap[-2] =r'$\bar{u}$'
fmap[-1] =r'$\bar{d}$'
fmap[-3] =r'$\bar{s}$'
nrows,ncols=3,3
fig = py.figure(figsize=(ncols*5,nrows*5))
AX={}
for i in range(9): AX[flavs[i]]=py.subplot(nrows,ncols,i+1)
measure=lambda A,B: np.abs((A-B)/A)*100
#measure=lambda A,B: (B/A)
nrep=50#len(true)
handler={}
flag=False
for i in range(nrep):
lprint('progress %d/%d'%(i+1,nrep))
t=load('%s/benchmark/%s'%(wdir,true[i]))
h=load('%s/benchmark/%s'%(wdir,lha[i]))
if flag==False:
for i in range(9):
for Q2 in sorted(t):
flav=flavs[i]
handler[Q2],=AX[flav].plot(X,t[Q2][flav])#,alpha=0.1)
#AX[flav].plot(X,h[Q2][flav],color=handler[Q2].get_color(),ls='--')
break
flag=True
else:
for i in range(9):
for Q2 in sorted(t):
flav=flavs[i]
c=handler[Q2].get_color()
AX[flav].plot(X,t[Q2][flav],color=handler[Q2].get_color())#,alpha=0.1)
#AX[flav].plot(X,h[Q2][flav],color=handler[Q2].get_color(),ls='--')
break
print
#AX[4].legend([Line2D([0,1], [0,0], linewidth=10
# , linestyle='-', color=handler[Q2].get_color())
# for Q2 in sorted(handler)]\
# ,[r'$Q^2=%0.1f$'%Q2 for Q2 in sorted(handler)]\
# ,fontsize=30,loc=2,bbox_to_anchor=(-1.1,-0.25),ncol=3)
AX[3].set_ylabel(r'$\rm rel.~err.(\%)$',size=40)
AX[4].set_xlabel(r'$x$',size=40)
AX[21].set_title(r'$\rm %s$'%(title).replace(' ',r'~'),size=40)
for _ in AX:
if xscale=='log':
AX[_].semilogx()
AX[_].set_xticks([1e-5,1e-3,1e-1])
#else:
# AX[_].set_xticks([0.1])
#AX[_].semilogy()
AX[_].text(0.1,0.8,r'$%s$'%fmap[_],transform=AX[_].transAxes,size=30)
#AX[_].set_ylim(1e-4,1)
AX[_].set_ylim(0,1)
#AX[_].set_ylim(-2,2)
AX[_].set_xlim(np.amin(X),1)
AX[_].tick_params(axis='both', which='major', labelsize=20)
#AX[_].set_yticks([1e-3,1e-2,1e-1])
AX[_].axhline(1e-1,ls=':',color='k')
checkdir('%s/gallery'%wdir)
#py.tight_layout()
py.subplots_adjust(left=0.1
,bottom=0.2
,right=0.95
,top=0.92
,wspace=None
,hspace=None)
py.savefig('%s/gallery/benchmark2-%s.jpg'%(wdir,dist))
py.close()
def gen_xf(wdir, flavors=['g', 'u', 'ub', 'd', 'db', 's', 'sb'], Q2=None):
load_config('%s/input.py' % wdir)
istep = core.get_istep()
replicas = core.get_replicas(wdir)
## 'conf' will be modified for each replica individually later in the loop over 'replicas'
## the reason for doing this is that 'fix parameters' has to be set correctly for each replica
if 'pdf' not in conf['steps'][istep]['active distributions']:
print('pdf-proton not in active distribution')
return
resman = RESMAN(nworkers=1, parallel=False, datasets=False)
parman = resman.parman
parman.order = replicas[0]['order'][istep]
## make sure 'parman' uses the same order for active distributions as all the replicas do
# print parman.order
pdf = conf['pdf']
## setup kinematics
xs = 10.0**np.linspace(-3, -1, 100)
xs = np.append(xs, np.linspace(0.1, 0.99, 100))
if Q2 == None: Q2 = conf['Q20']
print('\ngenerating pdf-proton from %s at Q2 = %.2f' % (wdir, Q2))
## compute xf for all replicas
xfs = {}
n_replicas = len(replicas)
for i in range(n_replicas):
lprint('%d/%d' % (i + 1, n_replicas))
## filter
#flag=False
#params=replica['params'][istep]
#order=replica['order'][istep]
#for i in range(len(order)):
# if order[i][0]!=1:continue
# if order[i][1]!='pdf':continue
# #if order[i][2]=='s1 a':
# # if params[i]<-0.9: flag=True
#if flag: continue
parman.order = copy.copy(replicas[i]['order'][istep])
core.mod_conf(istep, replicas[i])
parman.set_new_params(replicas[i]['params'][istep], initial=True)
for flavor in flavors:
if flavor not in xfs: xfs[flavor] = []
if flavor == 'rs':
func = lambda x: (pdf.get_xF(x, Q2, 's') + pdf.get_xF(
x, Q2, 'sb')) / (pdf.get_xF(x, Q2, 'db') + pdf.get_xF(
x, Q2, 'ub'))
elif flavor == 'uv':
func = lambda x: pdf.get_xF(x, Q2, 'u') - pdf.get_xF(
x, Q2, 'ub')
elif flavor == 'dv':
func = lambda x: pdf.get_xF(x, Q2, 'd') - pdf.get_xF(
x, Q2, 'db')
elif flavor == 'd/u':
func = lambda x: pdf.get_xF(x, Q2, 'd') / pdf.get_xF(
x, Q2, 'u')
elif flavor == 'db+ub':
func = lambda x: pdf.get_xF(x, Q2, 'db') + pdf.get_xF(
x, Q2, 'ub')
elif flavor == 'db-ub':
func = lambda x: pdf.get_xF(x, Q2, 'db') - pdf.get_xF(
x, Q2, 'ub')
elif flavor == 's+sb':
func = lambda x: pdf.get_xF(x, Q2, 's') + pdf.get_xF(
x, Q2, 'sb')
else:
func = lambda x: pdf.get_xF(x, Q2, flavor)
xfs[flavor].append(np.array([func(x) for x in xs]))
print
checkdir('%s/data' % wdir)
if Q2 == conf['Q20']:
save({
'X': xs,
'Q2': Q2,
'XF': xfs
}, '%s/data/pdf-%d.dat' % (wdir, istep))
else:
save({
'X': xs,
'Q2': Q2,
'XF': xfs
}, '%s/data/pdf-%d-%f.dat' % (wdir, istep, Q2))
