When I try and start sampling, the sampling never actually starts and I’m not sure why because an error isn’t being thrown.
python 3.8.8
pymc3 3.11.3
with pm.Model() as weibull_time:
beta_offense = pm.Normal('offense',mu=0.0,sigma=5,shape=total_teams)
beta_defense = pm.Normal('defense',mu=0.0,sigma=5,shape=total_teams)
beta_ha = pm.HalfCauchy('home_adv',beta=.5,shape=total_teams)
beta_gd = pm.Normal('goal_difference',mu=0.0,sigma=5,shape=total_gd)
beta_og = pm.Normal('off_goals',mu=0.0,sigma=5,shape=total_og)
beta_dg = pm.Normal('def_goals',mu=0.0,sigma=5,shape=total_dg)
s = pm.HalfNormal("s", 5.0)
mu_chance = beta_offense[home_adv_data_uncensored['offense'].values] + beta_defense[home_adv_data_uncensored['defense'].values]\
+ beta_ha[home_adv_data_uncensored['home_adv'].values] + beta_gd[home_adv_data_uncensored['goal_difference'].values]\
+ beta_og[home_adv_data_uncensored['rolling_off_goal_count'].values] + beta_dg[home_adv_data_uncensored['rolling_def_goal_count'].values]
y_obs = pm.Gumbel('y_obs',mu=mu_chance,beta=s,observed=home_adv_data_uncensored['survival_std'].values)
with weibull_time:
trace = pm.sample(tune=1000, cores=1, chains=1)
Any ideas as to why this would fail to even start? I’ve been using Bayesian Parametric Survival Analysis with PyMC3 — PyMC3 3.10.0 documentation as a rough guide for this problem