def get_p(i):
data, sim_scale = simulate_data(K, J, L, C, beta=beta)
data['Subj'] = pd.factorize(data['Subj'])[0] + 1
myThurst = thurstonian(design_formula='~0+Condition',
data=data,
subject_name="Subj")
return myThurst.sra_p_value(level_dict={'Condition': 'C2'})[0]
import numpy as np
import pandas as pd
from multiprocessing import Process
if __name__ == '__main__':
#Load and preprocess data
K = 3 #Number of items
J = 10 #Number of participants
L = 1 #Number of trials
C = 2 #Number of conditions
beta = np.array([[0.0, 1.0, 2.0], [0.0, 0.1, 0.2]])
data, sim_scale = simulate_data(K, J, L, C, beta=beta, seed=4354356)
data['Subj'] = pd.factorize(data['Subj'])[0] + 1
myThurst = thurstonian(design_formula='~0+Condition',
data=data,
subject_name="Subj")
myThurst.pre_sample()
P = [Process(target=run_sample, kwargs={'temp_fnum': i}) for i in range(4)]
for p in P:
p.start()
for p in P:
p.join()
@signal_base.function
def tm_prior(weights):
return weights * 10**40
# base par and tim file
parfile = 'J1713+0747.par'
timfile = 'J1713+0747.tim'
thetas = [0.05, 0.1, 0.15]
for theta in thetas:
# simulate data
idx = random.getrandbits(32)
sigma_out = 1e-6
simulate_data(parfile, timfile, theta=theta, idx=idx, sigma_out=sigma_out)
# grab simulated data
simparfile = 'simulated_data/outlier/{}/{}/J1713+0747.par'.format(
theta, idx)
simtimfile = 'simulated_data/outlier/{}/{}/J1713+0747.tim'.format(
theta, idx)
psr = Pulsar(simparfile, simtimfile)
simparfile = 'simulated_data/no_outlier/{}/{}/J1713+0747.par'.format(
theta, idx)
simtimfile = 'simulated_data/no_outlier/{}/{}/J1713+0747.tim'.format(
theta, idx)
psr2 = Pulsar(simparfile, simtimfile)
psrs = [psr, psr2]
#-------------------------------------#
#-------------------------------------#
#######################################
N = 5000
#######################################
#-------------------------------------#
#----------Generate test data---------#
#-------------------------------------#
#-------------------------------------#
#-------------------------------------#
#######################################
beta = np.array([[0.0, 1.0, 2.0], [0.0, -0.1, -0.2]])
data = simulate_data(K, J, L, C, beta=beta, seed=8547346)
data['Subj'] = pd.factorize(data['Subj'])[0] + 1
y = data[['Y1', 'Y2', 'Y3']].values
X = np.asarray(patsy.dmatrix('~Condition', data=data))
subj = data['Subj'].values
stan_data = dict(N=y.shape[0],
K=y.shape[1],
C=X.shape[1],
J=J,
rater=subj,
X=X,
beta_sd_prior=K * 3,
scale_sd_prior=0.5)
#Generate all data simulations
fit = gen.sampling(data=stan_data,
return np.sum(A(x)) / x.shape[0]
#Load and preprocess data
K = 5 #Number of items
J = 30 #Number of participants
L = 1 #Number of trials
C = 1 #Number of conditions
beta = np.array([[0, 0.5, 1]])
all_sra = []
rand_sra = []
for i in range(1000):
data = simulate_data(K, J, L, C, beta=beta)
y = data[['Y1', 'Y2', 'Y3']]
y.to_csv("y_test.csv", index=False)
y = y.values
y_unif = np.array([np.random.permutation(range(3)) + 1 for i in range(J)])
all_sra.append(np.sqrt(sra(y)))
rand_sra.append(np.sqrt(sra(y_unif)))
plt.hist(np.sqrt(rand_sra), color='r', alpha=0.25, normed=True)
plt.hist(np.sqrt(all_sra), color='b', alpha=0.25, normed=True)
#s_l = S(y, 1, 0)
#Set plotting preferences
sns.set_context("paper")
sns.set_style("whitegrid")
output_folder = "article/Figures/"
#Load and preprocess data
K = 3 #Number of items
J = 5 #Number of participants
L = 1 #Number of trials
C = 2 #Number of conditions
beta = np.array([[0.0, 1.0, 2.0],
[0.0, 0.0, 0.0]])
data, sim_scale = simulate_data(K, J, L, C, beta = beta)
data['Subj'] = pd.factorize(data['Subj'])[0]+1
myThurst = thurstonian(design_formula = '~0+Condition', data = data, subject_name = "Subj")
#Note: When running the sampler use a cmdline console, not iPython from Spyder. Unfortunately this fails due to some quirks in Spyder
resample = True
if resample:
#Prepare for sampling using multiple cores
myThurst.pre_sample()
#Sample with multiprocessing
P = [Process(target=PyThurstonian.run_sample, args = (5000,), kwargs = {'temp_fnum': i}) for i in range(4)]
#Load and preprocess data
K = 15
J = 20 #Number of participants
L = 1 #Number of trials
C = 2 #Number of conditions
beta = [np.arange(0, 15), np.arange(0, 15)]
beta[1] = beta[1][::-1]
beta = np.array(beta) *0.1
# beta = np.zeros((C, K))
data = simulate_data(K, J, L, C, beta = beta, seed = 234254324)
data['Subj'] = pd.factorize(data['Subj'])[0]+1
myThurst = thurstonian(design_formula = '~Condition', data = data, subject_name = "Subj")
resample = True
if resample:
myThurst.sample(iter = 5000, chains = 1) #Sample from the posterior
myThurst.save_samples('MCMC/complex_samples') #Save the samples for later
else:
#Load old samples
myThurst.load_samples('MCMC/complex_samples')