engine.py

# Copyright (c) 2015-present, Facebook, Inc.
# All rights reserved.
"""
Train and eval functions used in main.py
"""
from utils.losses import DistillationLoss
from timm.utils import accuracy, ModelEma
from typing import Iterable, Optional
from timm.data import Mixup
from utils import utils
import torch
import time
import math
import sys
import pickle

def train_one_epoch(model: torch.nn.Module,criterion: DistillationLoss,
                    data_loader: Iterable, optimizer: torch.optim.Optimizer,
                    device: torch.device, epoch: int, loss_scaler, max_norm: float = 0,
                    model_ema: Optional[ModelEma] = None, mixup_fn: Optional[Mixup] = None,
                    set_training_mode=True, args = None):
    model.train(set_training_mode)
    metric_logger = utils.MetricLogger(delimiter="  ")
    metric_logger.add_meter('lr', utils.SmoothedValue(window_size=1, fmt='{value:.6f}'))
    header = 'Epoch: [{}]'.format(epoch)

    print_freq  = int(len(data_loader) / 20)
    for idx, (samples, targets_ori) in enumerate(metric_logger.log_every(data_loader, print_freq, header)):
        if args.debug and idx == 10:
            break
        start = time.time()
        samples = samples.to(device, non_blocking=True)
        targets = targets_ori.to(device, non_blocking=True)

        if mixup_fn is not None:
            samples, targets = mixup_fn(samples, targets)

        if args.use_bce:
            targets = targets.gt(0.0).type(targets.dtype)

        loss, loss_student, loss_teacher, loss_consistency = torch.tensor([0.0], device=device), torch.tensor([0.0], device=device), torch.tensor([0.0], device=device), torch.tensor([0.0], device=device)
        with torch.cuda.amp.autocast():
            outputs = model(samples)
            if args.task_type[0]:
                loss_student = criterion(samples, outputs["x_student"], targets) * args.task_weight[0]


            if args.task_type[1]:
                loss_teacher = criterion(samples, outputs["x_teacher"], targets) * args.task_weight[1]

            if args.task_type[2]:
                logits = torch.softmax(outputs["x_student"], dim=-1)
                confidence_score = logits[range(len(targets_ori)), targets_ori] + logits[range(len(targets_ori)), targets_ori.flip(0)] if args.mixup else logits[range(len(targets_ori)), targets_ori]
                confidence_pass = (confidence_score > args.task_type[2]).float()
                if confidence_pass.sum():
                    loss_consistency = (outputs["loss_consistency"] * confidence_pass).sum() / confidence_pass.sum() * args.task_weight[2]
                    metric_logger.update(confidence_score = confidence_score.mean().item())
                    metric_logger.update(confidence_pass  = confidence_pass.mean().item())
            loss = loss_student + loss_teacher + loss_consistency

        loss_value = loss.item()

        if not math.isfinite(loss_value):
            print("Loss is {}, stopping training".format(loss_value))
            sys.exit(1)

        optimizer.zero_grad()
        # this attribute is added by timm on one optimizer (adahessian)
        is_second_order = hasattr(optimizer, 'is_second_order') and optimizer.is_second_order
        loss_scaler(loss, optimizer, clip_grad=max_norm,
                    parameters=model.parameters(), create_graph=is_second_order)

        torch.cuda.synchronize()
        if model_ema is not None:
            model_ema.update(model)

        metric_logger.update(lr=optimizer.param_groups[0]["lr"])
        metric_logger.update(time=time.time() - start)
        metric_logger.update(loss_student=loss_student.item())
        metric_logger.update(loss_teacher=loss_teacher.item())
        metric_logger.update(loss_cls_consis=loss_consistency.item())

    # gather the stats from all processes
    metric_logger.synchronize_between_processes()
    print("Averaged stats:", metric_logger)
    return {k: meter.global_avg for k, meter in metric_logger.meters.items()}


@torch.no_grad()
def evaluate(data_loader, model, device, r_eval = None):

    criterion = torch.nn.CrossEntropyLoss()
    metric_logger = utils.MetricLogger(delimiter = "  ")
    header = 'Test:'

    # switch to evaluation mode
    model.eval()
    for samples, targets in metric_logger.log_every(data_loader, 10, header):
        samples = samples.to(device, non_blocking=True)
        targets = targets.to(device, non_blocking=True)

        # compute output
        with torch.cuda.amp.autocast():
            x = model(samples, r_eval=r_eval)
        loss = criterion(x, targets)
        acc1, acc5 = accuracy(x, targets, topk=(1, 5))

        batch_size = samples.shape[0]
        metric_logger.update(loss=loss.item())
        metric_logger.meters['acc1'].update(acc1.item(), n=batch_size)
        metric_logger.meters['acc5'].update(acc5.item(), n=batch_size)


    # gather the stats from all processes
    metric_logger.synchronize_between_processes()
    print('* Acc@1 {top1.global_avg:.3f} Acc@5 {top5.global_avg:.3f} loss {losses.global_avg:.3f}'
          .format(top1=metric_logger.acc1, top5=metric_logger.acc5, losses=metric_logger.loss))

    return {k: meter.global_avg for k, meter in metric_logger.meters.items()}