def run_epoch()

in ocr/paragraph_segmentation_dcnn.py [0:0]

• 22 lines of code
• 8 McCabe index (conditional complexity)

def run_epoch(e, network, dataloader, loss_function, trainer, log_dir, print_name, is_train):
    total_loss = nd.zeros(1, ctx)
    for i, (data, label) in enumerate(dataloader):
        data = data.as_in_context(ctx)
        label = label.as_in_context(ctx)
        
        with autograd.record(train_mode=is_train):
            output = network(data)
            loss_i = loss_function(output, label)
        if is_train:
            loss_i.backward()
            trainer.step(data.shape[0])

        total_loss += loss_i.mean()
        
        if e % send_image_every_n == 0 and e > 0 and i == 0:
            output_image = draw_box_on_image(output.asnumpy(), label.asnumpy(), data.asnumpy())
    epoch_loss = float(total_loss .asscalar())/len(dataloader)
    
    with SummaryWriter(logdir=log_dir, verbose=False, flush_secs=5) as sw:
        sw.add_scalar('loss', {print_name: epoch_loss}, global_step=e)
        if e % send_image_every_n == 0 and e > 0:
            output_image[output_image<0] = 0
            output_image[output_image>1] = 1
            sw.add_image('bb_{}_image'.format(print_name), output_image, global_step=e)
            
    return epoch_loss