def test_full_estimation(self):
n_confs = 1
total_noise = 0.5
noise_scale = total_noise / np.sqrt(n_confs)
flip = False
gen_data_params = GenDataParams(
n_samples=100,
e1_dist=['laplace'],
e1_std=3.0,
e2_dist=['laplace'],
e2_std=3.0,
f1_coef=[noise_scale for _ in range(n_confs)],
f2_coef=[noise_scale for _ in range(n_confs)],
conf_dist=[['all'] for _ in range(n_confs)],
fix_causality=True, # x1 -> x2 (b21 is non-zero)
seed=0)
data = gen_artificial_data(gen_data_params)
print('b_true = {}'.format(data['b']))
xs = data['xs']
print(xs)
# Flip causal direction
if flip:
xs = np.vstack((xs[:, 1], xs[:, 0])).T
print('{} (flipped)'.format(data['causality_true']))
else:
print('{} (not flipped)'.format(data['causality_true']))
wbic_infer_params = FullBayesInferParams(metric='wbic',
n_mc_samples=10000,
vb_seed=1,
ic_seed=1)
wbic_infer_result = _infer_causality_with_posterior(
xs, wbic_infer_params)
print("inferered causality: {}".format(wbic_infer_result['causality']))
print("WBIC of estimated causality model {} : {}".format(
wbic_infer_result['causality'], wbic_infer_result['metric']))
print("WBIC of reverese model: {}".format(
wbic_infer_result['metric_rev']))
def _infer_causality_wbic_bml(seed):
wbic_infer_params = FullBayesInferParams(metric='wbic',
n_mc_samples=10000,
vb_seed=1,
ic_seed=3)
gen_data_params.seed = seed
data = gen_artificial_data(gen_data_params)
xs = data['xs']
result = _infer_causality_with_posterior(xs, wbic_infer_params)
return result['causality']
for seed in range(1):
print(_infer_causality_wbic_bml(seed))
def _infer_causality_wbic_bml(seed):
wbic_infer_params = FullBayesInferParams(metric='wbic',
n_mc_samples=10000,
vb_seed=1,
ic_seed=3)
gen_data_params.seed = seed
data = gen_artificial_data(gen_data_params)
xs = data['xs']
result = _infer_causality_with_posterior(xs, wbic_infer_params)
return result['causality']
def _gen_artificial_data_csv(csv_file, gen_data_params):
data = gen_artificial_data(gen_data_params)
xs = data['xs']
if data['causality_true'] == [1, 2]:
header = 'x1_src,x2_dst'
else:
header = 'x1_dst,x2_src'
np.savetxt(csv_file, xs, fmt='%s', delimiter=',', header=header,
comments='')
print('Artificial data is generated and saved as %s' % csv_file)
def make_testdata(csv_file, gen_data_params):
"""Create CSV file including artificial data.
"""
data = gen_artificial_data(gen_data_params)
xs = data['xs'].astype('S20')
if data['causality_true'] == [1, 2]:
header = ['x1_src', 'x2_dst']
else:
header = ['x1_dst', 'x2_src']
csv_data = np.vstack((header, xs))
np.savetxt(csv_file, csv_data, fmt='%s', delimiter=',')
print('Made artificial data and saved as %s.' % csv_file)
def gen_artificial_data_given_cond(ix_trial, cond):
"""Generate artificial data for given conditions (parameters).
"""
# Set parameters for generating artificial data
n_confs = cond['n_confs']
gen_data_params = deepcopy(gen_data_params_default)
gen_data_params.n_samples = cond['n_samples']
gen_data_params.conf_dist = [['all'] for _ in range(n_confs)]
gen_data_params.e1_dist = [cond['data_noise_type']]
gen_data_params.e2_dist = [cond['data_noise_type']]
noise_scale = cond['totalnoise'] / np.sqrt(n_confs)
gen_data_params.f1_coef = [noise_scale for _ in range(n_confs)]
gen_data_params.f2_coef = [noise_scale for _ in range(n_confs)]
# Generate artificial data
gen_data_params.seed = ix_trial
data = gen_artificial_data(gen_data_params)
return data
def test_gen_artificial_data(plot=False, n_confounders=1):
"""Only check the function runs without error.
"""
gen_data_params = GenDataParams(
e1_dist=['uniform'],
f1_coef=['r2intervals' for _ in range(n_confounders)],
f2_coef=['r2intervals' for _ in range(n_confounders)],
conf_dist=[['all'] for _ in range(n_confounders)])
n_samples = gen_data_params.n_samples
data = gen_artificial_data(gen_data_params)
xs = data['xs']
confs = data['confs']
assert (data['xs'].shape == (n_samples, 2))
print('xs.shape = {}'.format(xs.shape))
print('std(x1) = {}'.format(np.std(xs[:, 0])))
print('std(x2) = {}'.format(np.std(xs[:, 1])))
print('std(conf1) = {}'.format(np.std(confs[:, 0])))
print('std(conf2) = {}'.format(np.std(confs[:, 1])))
print('')
pprint(vars(gen_data_params))
plt.figure()
plt.scatter(xs[:, 0], xs[:, 1])
plt.xlabel('x1')
plt.xlabel('x2')
plt.title('Observations')
plt.figure()
plt.hist(data['es'])
plt.title('Errors')
plt.figure()
plt.hist(data['confs'])
plt.title('Confounders')