def run_luxus_HMC(sigmaB2, lux_data, stan_model_file, plot_file_name, covariate_to_be_tested, N_outputsamples, N_chains, diagnostic_plots,testtype,test2cov,N_cytosines):
time_start=time.time()
fit = stan_cache(stan_model_file, data=lux_data, iter=N_outputsamples, chains=N_chains)
samples=fit.extract(permuted=True)
if testtype==1:
bf=calculate_savagedickey_kde_1d(numpy.zeros((1,)),sigmaB2*numpy.eye(1),samples['B'][:,covariate_to_be_tested])
else:
bf=calculate_savagedickey_kde_1d(numpy.zeros((1,)),sigmaB2*numpy.eye(1),samples['B'][:,covariate_to_be_tested]-samples['B'][:,test2cov])
time_end_full=time.time()
print(fit)
if diagnostic_plots==1:
if N_cytosines ==1:
vars_to_be_printed=['l','sigmaE2','sigmaR2','B']
else:
vars_to_be_printed=['l','sigmaE2','sigmaC2','sigmaR2','B']
fit.plot(vars_to_be_printed)
pyplot.tight_layout()
pyplot.savefig(plot_file_name)
sigmaR2_mean=numpy.mean(samples['sigmaR2'])
sigmaE2_mean=numpy.mean(samples['sigmaE2'])
if N_cytosines==1:
sigmaC2_mean=numpy.nan
else:
sigmaC2_mean=numpy.mean(samples['sigmaC2'])
runtime=time_end_full-time_start
return bf, runtime, sigmaR2_mean, sigmaC2_mean, sigmaE2_mean;
def run_luxus_HMC(sigmaB2, lux_data, diff, stan_file_name, BF_output_file,
name_to_BF_file, isWindow, N_cyt_in_analysis, cytosineIndex,
d_file, B_file, plot_file_name, modelNro, N_predictors):
#cytosineIndex is only needed when N_cyt_in_analysis=1, otherwise any value can be given.
#ModelNro: 2 means LuxUS, 3 means LuxUS sep
time_start = time.time()
fit = stan_cache(stan_file_name, data=lux_data, iter=1000, chains=4)
time_end = time.time()
samples = fit.extract(permuted=True)
numpy.savetxt("%s_%s_%s.txt" % (B_file, name_to_BF_file, cytosineIndex),
samples['B'][:, 1],
delimiter=',')
if isWindow == 1:
bf = calculate_savagedickey_kde_1d(numpy.zeros((1, )),
sigmaB2 * numpy.eye(1),
samples['B'][:, 1])
with open(BF_output_file, 'a+') as fw:
fw.write("%f\t%f\t%s\n" % (bf, diff, name_to_BF_file))
else:
for c_ind in range(0, N_cyt_in_analysis):
bf = calculate_savagedickey_kde_1d(
numpy.zeros((1, )), sigmaB2 * numpy.eye(1),
samples['B'][:, c_ind * N_predictors + 1])
if N_cyt_in_analysis == 1:
with open(BF_output_file, 'a+') as f:
f.write("%f\t%f\t%f\t%s\n" %
(bf, diff, cytosineIndex, name_to_BF_file))
else:
with open(BF_output_file, 'a+') as f:
f.write("%f\t%f\t%f\t%s\n" %
(bf, diff, c_ind, name_to_BF_file))
time_end_full = time.time()
print(fit)
print(samples.keys())
if modelNro == 2:
#LuxUS
vars_to_be_printed = ['l', 'sigmaE2', 'sigmaC2', 'sigmaR2', 'B']
fit.plot(vars_to_be_printed)
pyplot.tight_layout()
else:
#LuxUS sep
vars_to_be_printed = ['sigmaE2', 'sigmaR2', 'B']
fit.plot(vars_to_be_printed)
pyplot.tight_layout()
savefig(plot_file_name)
thetas = samples['theta'].copy()
theta_mean = numpy.mean(thetas, 1)
return theta_mean, time_start, time_end, time_end_full
def run_luxus_HMC(sigmaB2, lux_data, stan_model_file, plot_file_name,
covariate_to_be_tested, N_outputsamples, N_chains,
diagnostic_plots, testtype, test2cov, N_cytosines,
N_predictors):
time_start = time.time()
fit = stan_cache(stan_model_file,
data=lux_data,
iter=N_outputsamples,
chains=N_chains)
samples = fit.extract(permuted=True)
bf = numpy.zeros(N_cytosines)
if testtype == 1:
for c_ind in range(0, N_cytosines):
bf[c_ind] = calculate_savagedickey_kde_1d(
numpy.zeros((1, )), sigmaB2 * numpy.eye(1),
samples['B'][:, 2 * c_ind + covariate_to_be_tested])
else:
for c_ind in range(0, N_cytosines):
bf[c_ind] = calculate_savagedickey_kde_1d(
numpy.zeros((1, )), sigmaB2 * numpy.eye(1),
samples['B'][:, 2 * c_ind + covariate_to_be_tested] -
samples['B'][:, 2 * c_ind + test2cov])
time_end_full = time.time()
print(fit)
if diagnostic_plots == 1:
if N_cytosines == 1:
vars_to_be_printed = ['sigmaE2', 'sigmaR2', 'B']
else:
vars_to_be_printed = ['l', 'sigmaE2', 'sigmaR2', 'B', 'd']
fit.plot(vars_to_be_printed)
pyplot.tight_layout()
pyplot.savefig(plot_file_name)
sigmaR2_mean = numpy.mean(samples['sigmaR2'])
sigmaE2_mean = numpy.mean(samples['sigmaE2'])
theta_median = numpy.median(samples['theta'], axis=0)
theta_mean = numpy.mean(samples['theta'], axis=0)
if N_cytosines > 1:
d_median = numpy.median(samples['d'], axis=0)
d_mean = numpy.mean(samples['d'], axis=0)
else:
d_median = numpy.array([0])
d_mean = numpy.array([0])
runtime = time_end_full - time_start
return bf, runtime, sigmaR2_mean, sigmaE2_mean, d_median[::
N_predictors], theta_median, d_mean[::
N_predictors], theta_mean
def run_luxus_VI(lux_data, model_name, N_cytosines, N_predictors, N_replicates,
N_gradsamples, N_elbosamples, N_outputsamples,
temp_input_data_file_name, temp_output_file_name, sigmaB2,
isWindow, name_to_BF_file, BF_output_file, diff,
cytosineIndex, d_file, B_file):
time_start = time.time()
pystan.misc.stan_rdump(lux_data, temp_input_data_file_name)
subprocess.call(
"./%s variational grad_samples=%s elbo_samples=%s output_samples=%s data file=%s output file=%s diagnostic_file=%s"
% (model_name, N_gradsamples, N_elbosamples, N_outputsamples,
temp_input_data_file_name, temp_output_file_name, d_file),
shell=True)
time_end = time.time()
samples = csvIntoExtractDict(temp_output_file_name, 0)
if isWindow == 1:
samples_B = extractVariable1dim(samples, 'B', N_predictors,
N_outputsamples)
bf = calculate_savagedickey_kde_1d(numpy.zeros((1, )),
sigmaB2 * numpy.eye(1),
samples_B[:, 1])
with open(BF_output_file, 'a+') as fw:
fw.write("%f\t%f\t%s\n" % (bf, diff, name_to_BF_file))
else:
samples_B = extractVariable1dim(samples, 'B',
N_predictors * N_cytosines,
N_outputsamples)
for c_ind in range(0, N_cytosines):
bf = calculate_savagedickey_kde_1d(
numpy.zeros((1, )), sigmaB2 * numpy.eye(1),
samples_B[:, c_ind * N_predictors + 1])
with open(BF_output_file, 'a+') as f:
if N_cytosines == 1:
f.write("%f\t%f\t%f\t%s\n" %
(bf, diff, cytosineIndex, name_to_BF_file))
else:
f.write("%f\t%f\t%f\t%s\n" %
(bf, diff, c_ind, name_to_BF_file))
time_end_full = time.time()
numpy.savetxt("%s_%s_%s.txt" % (B_file, name_to_BF_file, cytosineIndex),
samples_B,
delimiter=',')
samples_theta = extractVariable1dim(samples, 'theta',
2 * N_replicates * N_cytosines,
N_outputsamples)
theta_mean = numpy.mean(samples_theta, 1)
subprocess.call("rm %s" % (temp_input_data_file_name), shell=True)
subprocess.call("rm %s" % (temp_output_file_name), shell=True)
return theta_mean, time_start, time_end, time_end_full
def run_luxus_VI(lux_data, model_name, N_gradsamples, N_elbosamples,
N_outputsamples, temp_input_data_file_name,
temp_output_file_name, sigmaB2, diagnostic_plots,
plot_file_name, N_predictors, N_reps, covariate_to_be_tested,
testtype, test2cov, N_cytosines):
time_start = time.time()
pystan.misc.stan_rdump(lux_data, temp_input_data_file_name)
subprocess.call(
"./%s variational grad_samples=%s elbo_samples=%s output_samples=%s data file=%s output file=%s"
% (model_name, N_gradsamples, N_elbosamples, N_outputsamples,
temp_input_data_file_name, temp_output_file_name),
shell=True)
samples = csvIntoExtractDict(temp_output_file_name, 0)
samples_B = extractVariable1dim(samples, 'B', N_predictors * N_cytosines,
N_outputsamples)
bf = numpy.zeros(N_cytosines)
if testtype == 1:
bf = calculate_savagedickey_kde_1d(
numpy.zeros((1, )), sigmaB2 * numpy.eye(1),
samples_B[:, covariate_to_be_tested])
else:
bf = calculate_savagedickey_kde_1d(
numpy.zeros((1, )), sigmaB2 * numpy.eye(1),
samples_B[:, covariate_to_be_tested] - samples_B[:, test2cov])
time_end_full = time.time()
samples_sigmaR2 = extractVariable1dim(samples, 'sigmaR2', 1,
N_outputsamples)
samples_sigmaE2 = extractVariable1dim(samples, 'sigmaE2', 1,
N_outputsamples)
samples_sigmaC2 = extractVariable1dim(samples, 'sigmaC2', 1,
N_outputsamples)
samples_theta = extractVariable1dim(samples, 'theta', N_cytosines * N_reps,
N_outputsamples)
sigmaR2_mean = numpy.mean(samples_sigmaR2)
sigmaE2_mean = numpy.mean(samples_sigmaE2)
theta_median = numpy.median(samples_theta, axis=0)
if diagnostic_plots == 1:
pyplot.subplot(5, 1, 1)
pyplot.hist(samples_sigmaR2, bins=20)
pyplot.title(r'$\sigma_R^2$')
pyplot.show()
pyplot.subplot(5, 1, 2)
pyplot.hist(samples_sigmaE2, bins=20)
pyplot.title(r'$\sigma_E^2$')
pyplot.show()
pyplot.subplot(5, 1, 3)
pyplot.hist(samples_sigmaC2, bins=20)
pyplot.title(r'$\sigma_C^2$')
pyplot.show()
pyplot.subplot(5, 1, 4)
for b_ind in range(0, N_predictors):
pyplot.hist(samples_B[b_ind], bins=20, label="b_%s" % (b_ind))
pyplot.legend(loc='upper right')
if N_cytosines > 1:
samples_l = extractVariable1dim(samples, 'l', 1, N_outputsamples)
pyplot.subplot(5, 1, 5)
pyplot.hist(samples_l, bins=20)
pyplot.title(r'$\ell$')
pyplot.show()
pyplot.title(r'$\mathbf{b}$')
pyplot.show()
pyplot.savefig(plot_file_name)
subprocess.call("rm %s" % (temp_input_data_file_name), shell=True)
subprocess.call("rm %s" % (temp_output_file_name), shell=True)
runtime = time_end_full - time_start
return bf, runtime, sigmaR2_mean, sigmaE2_mean, theta_median