# In[20]:
utility_term_factory = utility_term_estimation.UtilityAggregatorFactory()
# # Evaluation
# In[21]:
train_measures = evaluation.Measures(
x,
loss,
u,
sample_predictive_y,
optimal_h_bayes_estimator=optimal_h_bayes_estimator,
y_mask=training_mask,
GAIN_OPTIMAL_H_NUMERICALLY=GAIN_OPTIMAL_H_NUMERICALLY,
RISK_OPTIMAL_H_NUMERICALLY=RISK_OPTIMAL_H_NUMERICALLY,
EVAL_NSAMPLES_UTILITY_TERM_THETA=EVAL_NSAMPLES_UTILITY_TERM_THETA,
EVAL_NSAMPLES_UTILITY_TERM_Y=EVAL_NSAMPLES_UTILITY_TERM_Y,
EVAL_MAX_NITER=EVAL_MAX_NITER,
EVAL_SGD_PREC=EVAL_SGD_PREC)
test_measures = evaluation.Measures(
x,
loss,
u,
sample_predictive_y,
optimal_h_bayes_estimator=optimal_h_bayes_estimator,
y_mask=testing_mask,
GAIN_OPTIMAL_H_NUMERICALLY=GAIN_OPTIMAL_H_NUMERICALLY,
# In[182]:
utility_term_factory = utility_term_estimation.UtilityAggregatorFactory()
# # Evaluation
# In[183]:
measures = evaluation.Measures(
torch.tensor(schools_dat["y"], dtype=torch.float32),
loss,
u,
sample_predictive_y,
optimal_h_bayes_estimator=optimal_h_bayes_estimator,
y_mask=utility_term_mask, # all data points
GAIN_OPTIMAL_H_NUMERICALLY=GAIN_OPTIMAL_H_NUMERICALLY,
RISK_OPTIMAL_H_NUMERICALLY=RISK_OPTIMAL_H_NUMERICALLY,
EVAL_NSAMPLES_UTILITY_TERM_THETA=EVAL_NSAMPLES_UTILITY_TERM_THETA,
EVAL_NSAMPLES_UTILITY_TERM_Y=EVAL_NSAMPLES_UTILITY_TERM_Y,
EVAL_MAX_NITER=EVAL_MAX_NITER,
EVAL_SGD_PREC=EVAL_SGD_PREC)
# # VI
# In[136]:
# intialization
torch.manual_seed(SEED)
np.random.seed(SEED)
