9a4540b682074d3f84a2fc232cf2a832dc96bce7,src/aggregate_results.py,,latexify,#Any#Any#,92

Before Change

        for run_name in results["valid"].keys():
            for metric, scores in results["valid"][run_name].items():
                res_str = "%s & %s & " % (run_name, metric)
                metric_scores = [(k, v) for k, v in scores.items()]
                metric_scores.sort(key=lambda x: x[0])
                //res_str += "& ".join([average(scores[task]) for task in TASKS if task in scores])
                res_str += " & ".join(["%.3f" % average(scores[k]) for (k, _) in metric_scores])
                res_str += " \\\\\n"
                fh.write(res_str)

// aggregated results

After Change

        header += "\\hline\n"
        fh.write(header)
        for run_name in run_order:
            for metric in metric_order:
                if not len(results["valid"][run_name][metric]):
                    continue
                res_str = "%s & %s" % (run_name, metric)
                scores = results["valid"][run_name][metric]
                for task in task_order:
                    try:
                        res_str += " & %.3f" % average(scores[task])
                    except:
                        pdb.set_trace()
                res_str += " \\\\\n"
                fh.write(res_str)

// aggregated results
// structure is something like
// run name: results
//           stopping criteria : metric : scores
all_results = {"valid": defaultdict(lambda: defaultdict(lambda: defaultdict(list))),
        "test": defaultdict(lambda: defaultdict(lambda: defaultdict(list)))}

// aggregate results

Instances