ecd656578d31171683e7928477132e75348317d1,pyprob/model.py,Model,posterior,#Model#,87

Before Change

                len_str_num_traces = len(str(num_traces))
                print("Time spent  | Time remain.| Progress             | {} | Accepted|Smp reuse| Traces/sec".format("Trace".ljust(len_str_num_traces * 2 + 1)))
                prev_duration = 0
            for i in range(num_traces):
                if util._verbosity > 1:
                    duration = time.time() - time_start
                    if (duration - prev_duration > util._print_refresh_rate) or (i == num_traces - 1):
                        prev_duration = duration
                        traces_per_second = (i + 1) / duration
                        print("{} | {} | {} | {}/{} | {} | {} | {:,.2f}       ".format(util.days_hours_mins_secs_str(duration), util.days_hours_mins_secs_str((num_traces - i) / traces_per_second), util.progress_bar(i+1, num_traces), str(i+1).rjust(len_str_num_traces), num_traces, "{:,.2f}%".format(100 * (traces_accepted / (i + 1))).rjust(7), "{:,.2f}%".format(100 * samples_reused / max(1, samples_all)).rjust(7), traces_per_second), end="\r")
                        sys.stdout.flush()
                candidate_trace = next(self._trace_generator(trace_mode=TraceMode.POSTERIOR, inference_engine=inference_engine, metropolis_hastings_trace=current_trace, observe=observe, *args, **kwargs))
                log_acceptance_ratio = math.log(current_trace.length_controlled) - math.log(candidate_trace.length_controlled) + candidate_trace.log_prob_observed - current_trace.log_prob_observed
                for variable in candidate_trace.variables_controlled:
                    if variable.reused:
                        log_acceptance_ratio += torch.sum(variable.log_prob)
                        log_acceptance_ratio -= torch.sum(current_trace.variables_dict_address[variable.address].log_prob)
                        samples_reused += 1
                samples_all += candidate_trace.length_controlled

                if state._metropolis_hastings_site_transition_log_prob is None:
                    warnings.warn("Trace did not hit the Metropolis Hastings site, ensure that the model is deterministic except pyprob.sample calls")
                else:
                    log_acceptance_ratio += torch.sum(state._metropolis_hastings_site_transition_log_prob)

                // print(log_acceptance_ratio)
                if math.log(random.random()) < float(log_acceptance_ratio):
                    traces_accepted += 1
                    current_trace = candidate_trace
                // do thinning
                if i % thinning_steps == 0:
                    posterior.add(map_func(current_trace))

            if util._verbosity > 1:
                print()

            posterior.finalize()

After Change

                        sys.stdout.flush()
                candidate_trace = next(self._trace_generator(trace_mode=TraceMode.POSTERIOR, inference_engine=inference_engine, metropolis_hastings_trace=current_trace, observe=observe, *args, **kwargs))
                if filter is not None:
                    if not filter(candidate_trace):
                        continue
                log_acceptance_ratio = math.log(current_trace.length_controlled) - math.log(candidate_trace.length_controlled) + candidate_trace.log_prob_observed - current_trace.log_prob_observed
                for variable in candidate_trace.variables_controlled:
                    if variable.reused:
                        log_acceptance_ratio += torch.sum(variable.log_prob)

Instances