def main(args):
np.random.seed(args.beta_num)
sim = SumstatSimulation(args.sim_name)
arch = Architecture(sim.architecture)
d = Dataset(sim.dataset)
# sample the beta
beta = arch.draw_effect_sizes(sim.dataset, sim.h2g)[:, 0]
# compute noiseless phenotypes slice by slice
Y = np.zeros(d.N)
t0 = time()
for s in d.slices():
# X will be N x M
print(int(time() - t0), ': getting genotypes from file. SNPs', s)
X = d.get_standardized_genotypes(s)
print('computing phenotypes. SNPs', s)
Y += X.dot(beta[s[0]:s[1]])
del X
# normalize the Y and the beta to the desired heritability
normalization = np.std(Y) / np.sqrt(sim.h2g)
if normalization == 0: normalization = 1 # just in case we have some 0s...
Y /= normalization
beta /= normalization
# write the betas and the noiseless phenotypes
pickle.dump(beta, sim.beta_file(args.beta_num, 'wb'), 2)
pickle.dump(Y, sim.noiseless_Y_file(args.beta_num, 'wb'), 2)
def main(args):
np.random.seed(args.beta_num)
sim = SumstatSimulation(args.sim_name)
arch = Architecture(sim.architecture)
d = Dataset(sim.dataset)
# sample the beta
beta = arch.draw_effect_sizes(sim.dataset, sim.h2g)[:, 0]
# compute noiseless phenotypes slice by slice
Y = np.zeros(d.N)
t0 = time()
for s in d.slices():
# X will be N x M
print(int(time() - t0), ": getting genotypes from file. SNPs", s)
X = d.get_standardized_genotypes(s)
print("computing phenotypes. SNPs", s)
Y += X.dot(beta[s[0] : s[1]])
del X
# normalize the Y and the beta to the desired heritability
normalization = np.std(Y) / np.sqrt(sim.h2g)
if normalization == 0:
normalization = 1 # just in case we have some 0s...
Y /= normalization
beta /= normalization
# write the betas and the noiseless phenotypes
pickle.dump(beta, sim.beta_file(args.beta_num, "wb"), 2)
pickle.dump(Y, sim.noiseless_Y_file(args.beta_num, "wb"), 2)
def submit(args):
sim = SumstatSimulation(args.sim_name)
my_args = ["--sim_name", args.sim_name, "main", "--beta_num", "$LSB_JOBINDEX"]
outfilepath = sim.path() + ".sim_betas.%I.out"
bsub.submit(
["python", "-u", paths.code + "sim/sim_betas.py"] + my_args,
outfilepath,
jobname="simbetas[1-" + str(sim.num_betas) + "]",
memory_GB=5,
)
def submit(args):
sim = SumstatSimulation(args.sim_name)
my_args = [
'--sim_name', args.sim_name, 'main', '--beta_num', '$LSB_JOBINDEX'
]
outfilepath = sim.path() + \
'.sim_betas.%I.out'
bsub.submit(['python', '-u', paths.code + 'sim/sim_betas.py'] + my_args,
outfilepath,
jobname='simbetas[1-' + str(sim.num_betas) + ']',
memory_GB=5)
def main(args):
np.random.seed(args.beta_num + args.sample_num * 10000)
sim = SumstatSimulation(args.sim_name)
d = Dataset(sim.dataset)
pretty.print_namespace(sim)
print()
# read in noiseless phenotypes
Y = pickle.load(sim.noiseless_Y_file(args.beta_num))
# choose individuals and create ensemble of Ys
indices = np.random.choice(Y.shape[0], size=(sim.sample_size, ))
Y = Y[indices]
# compute how much noise to add
sigma2e = 1 - sim.h2g
print('adding noise. sigma2e =', sigma2e)
Y += np.sqrt(sigma2e) * np.random.randn(*Y.shape)
if sim.condition_on_covariates:
print('projecting covariates out of Y')
Y = d.project_out_covariates(Y, covariates=d.covariates[indices])
alphahat = np.zeros(d.M)
t0 = time()
def compute_sumstats_for_slice(s):
# X will be N x M
print(int(time() - t0), ': getting genotypes from file. SNPs', s)
X = d.get_standardized_genotypes(s)[indices]
if sim.condition_on_covariates:
print(int(time() - t0), ': projecting out covariates')
X = d.project_out_covariates(X, covariates=d.covariates[indices])
print(int(time() - t0), ': computing sumstats. SNPs', s)
alphahat[s[0]:s[1]] = X.T.dot(Y) / sim.sample_size
del X
map(compute_sumstats_for_slice, d.slices())
# write output
def write_output():
pickle.dump(indices,
sim.individuals_file(args.beta_num, args.sample_num, 'wb'),
2)
pickle.dump(Y, sim.noisy_Y_file(args.beta_num, args.sample_num, 'wb'),
2)
pickle.dump(alphahat,
sim.sumstats_file(args.beta_num, args.sample_num, 'wb'), 2)
write_output()
def main(args):
np.random.seed(args.beta_num + args.sample_num * 10000)
sim = SumstatSimulation(args.sim_name)
d = Dataset(sim.dataset)
pretty.print_namespace(sim); print()
# read in noiseless phenotypes
Y = pickle.load(sim.noiseless_Y_file(args.beta_num))
# choose individuals and create ensemble of Ys
indices = np.random.choice(Y.shape[0], size=(sim.sample_size,))
Y = Y[indices]
# compute how much noise to add
sigma2e = 1 - sim.h2g
print('adding noise. sigma2e =', sigma2e)
Y += np.sqrt(sigma2e) * np.random.randn(*Y.shape)
if sim.condition_on_covariates:
print('projecting covariates out of Y')
Y = d.project_out_covariates(Y, covariates=d.covariates[indices])
alphahat = np.zeros(d.M)
t0 = time()
def compute_sumstats_for_slice(s):
# X will be N x M
print(int(time() - t0), ': getting genotypes from file. SNPs', s)
X = d.get_standardized_genotypes(s)[indices]
if sim.condition_on_covariates:
print(int(time() - t0), ': projecting out covariates')
X = d.project_out_covariates(X, covariates=d.covariates[indices])
print(int(time() - t0), ': computing sumstats. SNPs', s)
alphahat[s[0]:s[1]] = X.T.dot(Y) / sim.sample_size
del X
map(compute_sumstats_for_slice, d.slices())
# write output
def write_output():
pickle.dump(indices, sim.individuals_file(
args.beta_num, args.sample_num, 'wb'), 2)
pickle.dump(Y, sim.noisy_Y_file(
args.beta_num, args.sample_num, 'wb'), 2)
pickle.dump(alphahat, sim.sumstats_file(
args.beta_num, args.sample_num, 'wb'), 2)
write_output()
def submit(args):
sim = SumstatSimulation(args.sim_name)
def submit_beta(beta_num):
my_args = [
'--sim_name', args.sim_name, 'main', '--beta_num',
str(beta_num), '--sample_num', '$LSB_JOBINDEX'
]
outfilepath = \
sim.path_to_beta(beta_num) + \
'.sim_sumstats.%I.out'
bsub.submit(
['python', '-u', paths.code + 'sim/sim_sumstats.py'] + my_args,
outfilepath,
jobname='simsumstats' + str(beta_num) + '[1-' +
str(sim.num_samples_per_beta) + ']',
# memory_GB=10.5)
memory_GB=13)
map(submit_beta, range(1, sim.num_betas + 1))
def run_on_batch(est, args):
sim = SumstatSimulation(args.sim_name)
pretty.print_namespace(sim); print()
print('batch=', args.batch_num)
print(est)
est.run_and_save_results(args.batch_num, sim)
def __init__(self, simulation_names):
self.simulations = []
for sim_name in simulation_names:
self.simulations.append(SumstatSimulation(sim_name))
