tf.reduce_sum(variation_prior.log_prob(tf.cast(variation,
tf.float32))), floatX)
ld = variation_d.logdens(variation)
tf.add_to_collection('logdensities', ld[tf.newaxis])
tf.add_to_collection('priors', pp)
tf.summary.histogram('variation', variation)
tf.summary.scalar('mean_variation', tf.reduce_mean(variation))
global_inf = DFlow([
NVPFlow(dim=(VAR_DIM * 2 + 1) * VAR_DIM,
name='flow_{}'.format(i),
aux_vars=variation[tf.newaxis]) for i in range(6)
],
init_sigma=0.01)
global_prior = Normal(None, sigma=1.).logdens(global_inf.output)
tf.add_to_collection('priors', global_prior)
tf.add_to_collection('logdensities', global_inf.logdens)
individ_variation_prior = Normal((VAR_DIM * 2 + 1) * VAR_DIM,
sigma=variation,
mu=global_inf.output[0])
models = []
indivs = {}
with tf.variable_scope(tf.get_variable_scope(),
dtype=floatX,
reuse=tf.AUTO_REUSE):
for country, data in country_data.items():
with tf.variable_scope(country):