def get_training_stats()

in pytorch_translate/evals.py [0:0]

• 73 lines of code
• 25 McCabe index (conditional complexity)

def get_training_stats(trainer):
    stats = OrderedDict()
    if trainer.get_meter("train_loss") is not None:
        avg = trainer.get_meter("train_loss").avg
        if avg is not None:
            stats["loss"] = f"{avg:.3f}"
    if trainer.get_meter("train_nll_loss").count > 0:
        nll_loss = trainer.get_meter("train_nll_loss").avg
        stats["nll_loss"] = f"{nll_loss:.3f}"
    else:
        nll_loss = trainer.get_meter("train_nll_loss").avg
    stats["ppl"] = get_perplexity(nll_loss) if nll_loss is not None else -1.0
    if trainer.get_meter("wps") is not None:
        stats["wps"] = (
            round(utils.item(trainer.get_meter("wps").avg))
            if trainer.get_meter("wps").avg
            else None
        )
    if trainer.get_meter("ups") is not None:
        stats["ups"] = (
            f"{trainer.get_meter('ups').avg:.1f}"
            if trainer.get_meter("ups").avg
            else None
        )
    if trainer.get_meter("wpb") is not None:
        stats["wpb"] = (
            round(utils.item(trainer.get_meter("wpb").avg))
            if trainer.get_meter("wpb").avg
            else None
        )
    if trainer.get_meter("bsz") is not None:
        stats["bsz"] = (
            round(utils.item(trainer.get_meter("bsz").avg))
            if trainer.get_meter("bsz").avg
            else None
        )
    stats["num_updates"] = trainer.get_num_updates()
    stats["lr"] = trainer.get_lr()
    if trainer.get_meter("gnorm") is not None:
        stats["gnorm"] = (
            f"{trainer.get_meter('gnorm').avg:.3f}"
            if trainer.get_meter("gnorm").avg
            else None
        )
    if trainer.get_meter("clip") is not None:
        stats["clip"] = (
            f"{trainer.get_meter('clip').avg:.0%}"
            if trainer.get_meter("clip").avg
            else None
        )
    if trainer.get_meter("oom") is not None:
        stats["oom"] = (
            trainer.get_meter("oom").avg if trainer.get_meter("oom").avg else None
        )
    if trainer.get_meter("loss_scale") is not None:
        stats["loss_scale"] = (
            f"{trainer.get_meter('loss_scale').avg:.3f}"
            if trainer.get_meter("loss_scale").avg
            else None
        )
    if trainer.get_meter("wall") is not None:
        stats["wall"] = (
            round(utils.item(trainer.get_meter("wall").elapsed_time))
            if trainer.get_meter("wall").elapsed_time
            else None
        )
    if trainer.get_meter("train_wall") is not None:
        stats["train_wall"] = (
            round(utils.item(trainer.get_meter("train_wall").sum))
            if trainer.get_meter("train_wall").sum
            else None
        )
    return stats