def simulation_iter(obs, nout, rho, variance, te, redraw, falsenull, ntreat=None, pmat=None, seed=None, n_jobs=1, simdata=None):
print 'Simulating for %i outcomes, %i false null hypotheses...' % (nout, falsenull)
# generate covariance matrix
i = [1 for j in range(0,nout)]
cov = np.outer(i, i) * rho
for j in range(0,nout):
cov[j,j] = variance
# generate matrix of means
mean = [0 for j in range(0,nout)]
data = genmultinorm(obs=obs, cov=cov, mean=mean, te=te, falsenull=falsenull, ntreat=ntreat)
# the first argument is regular permutation; the second argument is stepdown
perm, sd = permcombo(data=data, outcomes=['y{}'.format(j) for j in range(nout)], d='d' , redraw=redraw, seed=seed, pmat=pmat,
quiet=True, n_jobs=1, stepdown=True)
sd1t = sd.loc[:,'stepdown_1t'].values
sd1t.sort()
sd2t = sd.loc[:,'stepdown_2t'].values
sd2t.sort()
#return sd1t, sd2t
return_data = pd.DataFrame([sd1t, sd2t], index=['1t', '2t'])
return return_data
variance = 1 # variance/cov. matrix diagonal
obs = 50 # number of observations
ntreat = 25 # number of people assigned to treatment
te = 3 # treatment effect/effect size
redraw = 1000 # number of permutations
# generate covariance matrix
i = [1 for j in range(0,nout)]
cov = np.outer(i, i) * rho
for j in range(0,nout):
cov[j,j] = variance
# generate matrix of means
mean = [0 for j in range(0,nout)]
data = genmultinorm(obs=obs, cov=cov, mean=mean, te=te, ntreat = ntreat, falsenull = falsenull)
# generate matrix of permutations
tmp_d = [1 for j in range(ntreat)] + [0 for k in range(obs-ntreat)]
pmat = pd.DataFrame(np.squeeze(np.array([np.random.permutation(tmp_d) for j in range(redraw)]).T))
#perm, tippett= permcombo(data=data, outcomes=['y{}'.format(j) for j in range(nout)], d='d' , redraw=redraw, tippett=True, posneg=False, pmat = pmat, quiet=False, n_jobs=1)
#perm, stepdown = permcombo(data=data, outcomes=['y{}'.format(j) for j in range(nout)], d='d' , redraw=redraw, quiet=False, stepdown=True, n_jobs=1)
start_time = time.time()
perm, step = permcombo(data=data, outcomes=['y{}'.format(j) for j in range(nout)], d='d' , redraw=redraw, quiet=False, n_jobs=1, pmat=pmat, stepdown=True)
print("--- %s seconds ---" % (time.time() - start_time))
# parameters for fake data
nout = 8 # number of outcomes
falsenull = 8 # number of false nulls
rho = 0.0 # equicorrelation parameter
variance = 1 # variance/cov. matrix diagonal
obs = 56 # number of observations
ntreat = 28 # number of people assigned to treatment
te = 0.5 # treatment effect/effect size
redraw = 100 # number of permutations
# generate covariance matrix
i = [1 for j in range(0,nout)]
cov = np.outer(i, i) * rho
for j in range(0,nout):
cov[j,j] = variance
# generate matrix of means
mean = [0 for j in range(0,nout)]
data = genmultinorm(obs=obs, cov=cov, mean=mean, te=te, falsenull = falsenull, seed=None)
# generate matrix of permutations
tmp_d = [1 for j in range(ntreat)] + [0 for k in range(obs-ntreat)]
pmat = pd.DataFrame(np.squeeze(np.array([np.random.permutation(tmp_d) for j in range(redraw)]).T))
perm = permfactor(data=data, outcomes=['y{}'.format(j) for j in range(nout)], d='d' , redraw=redraw, pmat = pmat, quiet=False, n_jobs=1)
