Error when setting enable_aks=True in version 1.2.0 #1042
Description
When I set enable_aks=True in my ModelSpec, the model allows me to generate the instanced object, however when I sample from either prior or posterior distributions I get an error :
459 population_scaled_mean = np.mean(population_scaled_kpi)
460 population_scaled_stdev = np.std(population_scaled_kpi)
_UFuncTypeError: Cannot cast ufunc 'divide' output from dtype('float64') to dtype('int64') with casting rule 'same_kind'_
My previous set up is as follows:
prior = prior_distribution.PriorDistribution(
contribution_m=tfp.distributions.Beta(scaled_spends, 100 - scaled_spends),
gamma_c = tfp.distributions.Normal(0, 5), # Default, broad uninformative (used for seasonality control)
alpha_m = tfp.distributions.Beta(alpha_m_alpha, alpha_m_beta)
)
model_spec = spec.ModelSpec(
#knots = 8,
enable_aks=True,
max_lag = 12, # elongate from default 8 weeks, but control using alpha_m
media_prior_type = 'contribution',
prior=prior)
mmm = model.Meridian(input_data=data, model_spec=model_spec)
and my calls (which error)
mmm.sample_prior(500)
mmm.sample_posterior(n_chains=4, n_adapt=500, n_burnin=500, n_keep=1000)
I'm running a national model.
If I comment out enable_aks and uncomment knots=8 then the sampling proceeds as normal.