def simSuite(taus=[2.0],harms=[3],periods=[1.5e-3],snrs=[10],
maxfun = 5):
np.random.seed(42) # seed the random number generator - Dan - 2012-11-18
for tau in taus:
for harm in harms:
for period in periods:
for snr in snrs:
prof = makeProfile(harm)
ht = makeht(tau, nlag=2048)
CS = initSim(ht,prof,ref_freq=1/period,bw=1.0,noise=snr,source = ('tau_%.1f_nharm_%d_period_%.2f_snr_%.3f' % (tau,harm,period*1000.,snr)))
CS.initProfile()
CS.pp_meas = CS.pp_ref.copy()
CS.pp_ref = prof
CS.ph_ref = pycyc.phase2harm(CS.pp_ref)
CS.ph_ref = pycyc.normalize_profile(CS.ph_ref)
CS.ph_ref[0] = 0
CS.s0 = CS.ph_ref.copy()
CS.plotdir = '.'
CS.pharm = harm
CS.tau = tau
CS.noise = snr
CS.loop(make_plots=False,maxfun=maxfun)
CS.noise = snr
CS.saveState(os.path.join(CS.plotdir,'cs_' + CS.source + '.pkl'))
pycyc.plotSimulation(CS)
def simSuite(taus=[2.0], harms=[3], periods=[1.5e-3], snrs=[10], maxfun=5):
np.random.seed(42) # seed the random number generator - Dan - 2012-11-18
for tau in taus:
for harm in harms:
for period in periods:
for snr in snrs:
prof = makeProfile(harm)
ht = makeht(tau, nlag=2048)
CS = initSim(
ht,
prof,
ref_freq=1 / period,
bw=1.0,
noise=snr,
source=('tau_%.1f_nharm_%d_period_%.2f_snr_%.3f' %
(tau, harm, period * 1000., snr)))
CS.initProfile()
CS.pp_meas = CS.pp_ref.copy()
CS.pp_ref = prof
CS.ph_ref = pycyc.phase2harm(CS.pp_ref)
CS.ph_ref = pycyc.normalize_profile(CS.ph_ref)
CS.ph_ref[0] = 0
CS.s0 = CS.ph_ref.copy()
CS.plotdir = '.'
CS.pharm = harm
CS.tau = tau
CS.noise = snr
CS.loop(make_plots=False, maxfun=maxfun)
CS.noise = snr
CS.saveState(
os.path.join(CS.plotdir, 'cs_' + CS.source + '.pkl'))
pycyc.plotSimulation(CS)
def simSuite(taus=[2.0,10.0,100.0],harms=[3,10],periods=[1.5e-3,4e-3,10e-3],snrs=[10,100],
maxfun = 100):
for tau in taus:
for harm in harms:
for period in periods:
for snr in snrs:
prof = makeProfile(harm)
ht = makeht(tau, nlag=2048)
CS = initSim(ht,prof,ref_freq=1/period,bw=1.0,noise=snr,source = ('tau_%.1f_nharm_%d_period_%.2f_snr_%.3f' % (tau,harm,period*1000.,snr)))
CS.initProfile()
CS.pp_meas = CS.pp_ref.copy()
CS.pp_ref = prof
CS.ph_ref = pycyc.phase2harm(CS.pp_ref)
CS.ph_ref = pycyc.normalize_profile(CS.ph_ref)
CS.ph_ref[0] = 0
CS.s0 = CS.ph_ref.copy()
CS.plotdir = '/home/gjones/workspace/pycyc/simsuite'
CS.pharm = harm
CS.tau = tau
CS.noise = snr
CS.loop(make_plots=False,maxfun=maxfun)
CS.noise = snr
CS.saveState(os.path.join(CS.plotdir,'cs_' + CS.source + '.pkl'))
pycyc.plotSimulation(CS)
def initSim(ht,prof,ref_freq,bw,rf = None,filename= None,source='fake',noise = None):
if rf is None:
rf = np.abs(bw)/2.0
if ht.ndim == 1:
ht = ht[None,:]
CS = pycyc.CyclicSolver()
CS.ht0 = ht
if filename:
CS.filename = filename
else:
CS.filename = 'sim%s_%.1fMHz_%.1fms' % (source,bw,1000.0/ref_freq)
CS.nchan = ht.shape[1]
CS.nlag = CS.nchan
CS.nphase = prof.shape[0]
CS.nbin = CS.nphase
CS.nharm = CS.nphase/2 + 1
CS.source = source
CS.nspec = ht.shape[0]
CS.dynamic_spectrum = np.zeros((CS.nspec,CS.nchan))
CS.optimized_filters = np.zeros((CS.nspec,CS.nchan),dtype='complex')
CS.intrinsic_profiles = np.zeros((CS.nspec,CS.nbin))
CS.nopt = 0
CS.nloop = 0
CS.nopt = 0
CS.ref_freq = ref_freq
CS.bw = bw
CS.ref_phase = 0
CS.rf = rf
CS.hf_prev = np.ones((CS.nchan,),dtype='complex')
CS.pp_int = np.zeros((CS.nphase)) #intrinsic profile
CS.pp_ref = prof.copy()
CS.ph_ref = pycyc.phase2harm(CS.pp_ref)
CS.ph_ref = pycyc.normalize_profile(CS.ph_ref)
CS.ph_ref[0] = 0
CS.s0 = CS.ph_ref.copy()
CS.data = np.empty((CS.nspec,1,CS.nchan,CS.nphase), dtype='complex')
for k in range(CS.nspec):
CS.data[k,0,:,:] = pycyc.cs2ps(CS.modelCS(ht[k,:]))
if noise is not None:
signal = np.abs(CS.data).sum()
CS.noise = noise
rn = (np.random.randn(CS.data.shape[0],CS.data.shape[1],CS.data.shape[2],CS.data.shape[3]).astype('complex')
+1j*np.random.randn(CS.data.shape[0],CS.data.shape[1],CS.data.shape[2],CS.data.shape[3]))
rnpow = np.abs(rn).sum()
fact = signal/rnpow
CS.data = CS.data + rn*(fact/noise)
return CS
def initSim(ht,prof=std_prof,ref_freq=ref_freq,bw=bw,noise = None):
CS = pycyc.CyclicSolver()
CS.ht0 = ht
CS.filename = 'sim1643'
CS.nchan = ht.shape[0]
CS.nlag = CS.nchan
CS.nphase = prof.shape[0]
CS.nharm = CS.nphase/2 + 1
CS.source = 'J1643-1224sim'
CS.nspec = 1
CS.dynamic_spectrum = np.zeros((CS.nspec,CS.nchan))
CS.optimized_filters = np.zeros((CS.nspec,CS.nchan),dtype='complex')
CS.nopt = 0
CS.ref_freq = ref_freq
CS.bw = bw
CS.ref_phase = 0
CS.rf = rf
CS.hf_prev = np.ones((CS.nchan,),dtype='complex')
CS.pp_int = np.zeros((CS.nphase)) #intrinsic profile
CS.pp_ref = prof.copy()
CS.ph_ref = pycyc.phase2harm(CS.pp_ref)
CS.ph_ref = pycyc.normalize_profile(CS.ph_ref)
CS.ph_ref[0] = 0
CS.s0 = CS.ph_ref.copy()
csm = CS.modelCS(ht)
CS.data = pycyc.cs2ps(csm)[None,None,:,:]
CS.cs = csm.copy()
if noise is not None:
fact = CS.data.std()*noise
CS.data += (np.random.randn(CS.data.shape[2],CS.data.shape[3]).astype('complex')+1j*np.random.randn(CS.data.shape[2],CS.data.shape[3]))*fact
return CS
hf = np.loadtxt(os.path.join(dirname,'hf.tst'), delimiter=',').view('complex').squeeze()
hf_prev = np.loadtxt(os.path.join(dirname,'hf_prev.tst'), delimiter=',').view('complex').squeeze()
pp = np.loadtxt(os.path.join(dirname,'pp.tst'), delimiter=',')
pp2 = np.loadtxt(os.path.join(dirname,'pp2pp.tst'), delimiter=',')
pp_ref = np.loadtxt(os.path.join(dirname,'pp_ref.tst'), delimiter=',')
ph = np.loadtxt(os.path.join(dirname,'ph.tst'), delimiter=',').view('complex').squeeze()
ph_ref = np.loadtxt(os.path.join(dirname,'ph_ref.tst'), delimiter=',').view('complex').squeeze()
ph_recon = np.loadtxt(os.path.join(dirname,'ph_reconstructed.tst'), delimiter=',').view('complex').squeeze()
grad1 = np.loadtxt(os.path.join(dirname,'grad1.tst'), delimiter=',')
grad2 = np.loadtxt(os.path.join(dirname,'grad2.tst'), delimiter=',')
ntests = 0
nfail = 0
ntests += 1
phpperr = np.abs(ph - pycyc.phase2harm(pp)).mean()
print "(Cyclic-Modelling vs pycyc) phase2harm error:", phpperr
if phpperr > 1e-7:
print "phase2harm does not agree"
nfail += 1
ntests += 1
pp2pperr = np.abs(pp-pp2).mean()
print "(Cyclic-Modelling) test of harm2phase(phase2harm) == identity: ", pp2pperr
if pp2pperr > 1e-7:
print "error in cyclic_utils.c: harm2phase is not exact inverse of phase2harm"
nfail +=1
ntests += 1
phreconerr = np.abs(ph[1:]-ph_recon[1:]).mean() #skip zeroth harmonic since it's zeroed in ph
print "(Cyclic-Modelling) test if ph reconstructed from model cs and given hf matches original ph: ", phreconerr
# -*- coding: utf-8 -*-
"""
Created on Sat Jan 05 15:08:30 2013
@author: Glenn
"""
import simcyc
import pycyc
import numpy as np
from matplotlib import pyplot as plt
pp = np.fft.fftshift(simcyc.makeProfile(20))
ph = pycyc.phase2harm(pp)
"""
f = plt.figure()
ax = f.add_subplot(111)
ax.semilogx(20*np.log10(np.abs(ph)/np.abs(ph).max()))
ax.set_ylim(-55,5)
ax.set_xlim(1,300)
ax.grid(True)
ax.set_ylabel('20*log10(|p|)')
ax.set_xlabel('Harmonic')
f = plt.figure()
ax = f.add_subplot(111)
ax.plot(pp)
def initSim(ht,
prof,
ref_freq,
bw,
rf=None,
filename=None,
source='fake',
noise=None):
if rf is None:
rf = np.abs(bw) / 2.0
if ht.ndim == 1:
ht = ht[None, :]
CS = pycyc.CyclicSolver()
CS.ht0 = ht
if filename:
CS.filename = filename
else:
CS.filename = 'sim%s_%.1fMHz_%.1fms' % (source, bw, 1000.0 / ref_freq)
CS.nchan = ht.shape[1]
CS.nlag = CS.nchan
CS.nphase = prof.shape[0]
CS.nbin = CS.nphase
CS.nharm = CS.nphase / 2 + 1
CS.source = source
CS.nspec = ht.shape[0]
CS.dynamic_spectrum = np.zeros((CS.nspec, CS.nchan))
CS.optimized_filters = np.zeros((CS.nspec, CS.nchan), dtype='complex')
CS.intrinsic_profiles = np.zeros((CS.nspec, CS.nbin))
CS.nopt = 0
CS.nloop = 0
CS.nopt = 0
CS.ref_freq = ref_freq
CS.bw = bw
CS.ref_phase = 0
CS.rf = rf
CS.hf_prev = np.ones((CS.nchan, ), dtype='complex')
CS.pp_int = np.zeros((CS.nphase)) #intrinsic profile
CS.pp_ref = prof.copy()
CS.ph_ref = pycyc.phase2harm(CS.pp_ref)
CS.ph_ref = pycyc.normalize_profile(CS.ph_ref)
CS.ph_ref[0] = 0
CS.s0 = CS.ph_ref.copy()
CS.data = np.empty((CS.nspec, 1, CS.nchan, CS.nphase), dtype='complex')
for k in range(CS.nspec):
CS.data[k, 0, :, :] = pycyc.cs2ps(CS.modelCS(ht[k, :]))
if noise is not None:
signal = np.abs(CS.data).sum()
CS.noise = noise
rn = (np.random.randn(CS.data.shape[0], CS.data.shape[1],
CS.data.shape[2],
CS.data.shape[3]).astype('complex') +
1j * np.random.randn(CS.data.shape[0], CS.data.shape[1],
CS.data.shape[2], CS.data.shape[3]))
rnpow = np.abs(rn).sum()
fact = signal / rnpow
CS.data = CS.data + rn * (fact / noise)
return CS
