train.py

import os
import time
from tqdm import tqdm
import torch
import math

from torch.utils.data import DataLoader
from torch.nn import DataParallel

from nets.attention_model import set_decode_type
from utils.log_utils import log_values
from utils import move_to
from multiprocessing import Pool
import multiprocessing
import signal


def initializer():
    """Ignore CTRL+C in the worker process."""
    signal.signal(signal.SIGINT, signal.SIG_IGN)

def get_inner_model(model):
    return model.module if isinstance(model, DataParallel) else model


def validate(model, dataset, opts):
    # Validate
    print('Validating...')
    cost = rollout(model, dataset, opts)
    avg_cost = cost.mean()
    print('Validation overall avg_cost: {} +- {}'.format(
        avg_cost, torch.std(cost) / math.sqrt(len(cost))))

    return avg_cost


def rollout(model, dataset, opts):
    # Put in greedy evaluation mode!
    set_decode_type(model, "greedy")
    model.eval()

    def eval_model_bat(bat):
        with torch.no_grad():
            cost, _ = model(move_to(bat, opts.device))
        return cost.data.cpu()

    return torch.cat([
        eval_model_bat(bat)
        for bat
        in tqdm(DataLoader(dataset, batch_size=opts.eval_batch_size), disable=opts.no_progress_bar)
    ], 0)


def clip_grad_norms(param_groups, max_norm=math.inf):
    """
    Clips the norms for all param groups to max_norm and returns gradient norms before clipping
    :param optimizer:
    :param max_norm:
    :param gradient_norms_log:
    :return: grad_norms, clipped_grad_norms: list with (clipped) gradient norms per group
    """
    grad_norms = [
        torch.nn.utils.clip_grad_norm_(
            group['params'],
            max_norm if max_norm > 0 else math.inf,  # Inf so no clipping but still call to calc
            norm_type=2
        )
        for group in param_groups
    ]
    grad_norms_clipped = [min(g_norm, max_norm) for g_norm in grad_norms] if max_norm > 0 else grad_norms
    return grad_norms, grad_norms_clipped






def baseline_ls_async_generate(dataset, epoch_size):
    baseline_LS_name = 'bl_ls.txt'
    from LS import LS1
    import numpy as np
    def write_async(dataset):
        # np.random.seed(i)
        for loc in dataset:
            loc = loc.numpy()
            # print(loc.shape)
            tour = LS1.LS(loc, 7)
            cost = LS1.dist(loc, tour)
            print(cost)
            time.sleep(np.random.rand(1) * 2)
            f = open(baseline_LS_name, 'a')
            f.write(str(cost))
            f.write("\n")
            f.flush()
            f.close()
    pool = Pool(initializer=initializer)
    locs = [item['loc'] for item in dataset[:]]
    n_workers = multiprocessing.cpu_count()
    nums = int(epoch_size / n_workers)
    try:
        for i in range(n_workers):
            worker_dataset = locs[i * nums:(i + 1) * nums]
            pool.apply_async(write_async, args=(worker_dataset,))
    except KeyboardInterrupt:
        print("Stop!!!!!!!!!!!!")
        pool.terminate()

def train_epoch(model, optimizer, baseline, lr_scheduler, epoch, val_dataset, problem, tb_logger, opts):
    print("Start train epoch {}, lr={} for run {}".format(epoch, optimizer.param_groups[0]['lr'], opts.run_name))
    step = epoch * (opts.epoch_size // opts.batch_size)
    start_time = time.time()
    lr_scheduler.step(epoch)

    if not opts.no_tensorboard:
        tb_logger.log_value('learnrate_pg0', optimizer.param_groups[0]['lr'], step)

    # Generate new training data for each epoch

    if opts.test_instance is None:
        dataset = problem.make_dataset(
            size=opts.graph_size, num_samples=opts.epoch_size, distribution=opts.data_distribution)
    else:
        from test_plot.test_dataset import Test_CSPDataset
        dataset = Test_CSPDataset(size=opts.graph_size, num_samples=opts.epoch_size,
                                  cover_range=opts.cover, seed=1234,
                                  sample_mode=True)
        val_dataset = Test_CSPDataset(size=opts.graph_size, num_samples=opts.val_size,
                                  cover_range=opts.cover, seed=1234,
                                  sample_mode=True)
    training_dataset = baseline.wrap_dataset(dataset)
    training_dataloader = DataLoader(training_dataset, batch_size=opts.batch_size, num_workers=1)

    # baseline LS
    # pool.terminate()
    # baseline_ls_async_generate(dataset, opts.epoch_size)


    # Put model in train mode!
    model.train()
    set_decode_type(model, "sampling")

    for batch_id, batch in enumerate(tqdm(training_dataloader, disable=opts.no_progress_bar)):
        t1=time.time()
        train_batch(
            model,
            optimizer,
            baseline,
            epoch,
            batch_id,
            step,
            batch,
            tb_logger,
            opts
        )
        print("One batch:",time.time()-t1)
        step += 1

    epoch_duration = time.time() - start_time
    print("Finished epoch {}, took {} s".format(epoch, time.strftime('%H:%M:%S', time.gmtime(epoch_duration))))

    if (opts.checkpoint_epochs != 0 and epoch % opts.checkpoint_epochs == 0) or epoch == opts.n_epochs - 1:
        print('Saving model and state...')
        torch.save(
            {
                'model': get_inner_model(model).state_dict(),
                'optimizer': optimizer.state_dict(),
                'rng_state': torch.get_rng_state(),
                'cuda_rng_state': torch.cuda.get_rng_state_all(),
                'baseline': baseline.state_dict()
            },
            os.path.join(opts.save_dir, 'epoch-{}.pt'.format(epoch))
        )

    avg_reward = validate(model, val_dataset, opts)

    if not opts.no_tensorboard:
        tb_logger.log_value('val_avg_reward', avg_reward, step)

    baseline.epoch_callback(model, epoch)


def train_batch(
        model,
        optimizer,
        baseline,
        epoch,
        batch_id,
        step,
        batch,
        tb_logger,
        opts
):
    x, bl_val = baseline.unwrap_batch(batch)
    x = move_to(x, opts.device)
    bl_val = move_to(bl_val, opts.device) if bl_val is not None else None

    # Evaluate model, get costs and log probabilities
    cost, log_likelihood = model(x)

    # Evaluate baseline, get baseline loss if any (only for critic)
    bl_val, bl_loss = baseline.eval(x, cost) if bl_val is None else (bl_val, 0)

    # Baseline LS
    # import os
    # import numpy as np
    # if os.path.exists('bl_ls.txt'):
    #     bs_costs = np.loadtxt('bl_ls.txt')
    #     bl_val = bs_costs.mean()
    # else:
    #     bl_val = 0
    # print("Current basline_LS: ", bl_val)
    # bl_loss = 0
    # bl_val = bl_val * 0.8
    # Calculate loss
    reinforce_loss = ((cost - bl_val) * log_likelihood).mean()
    loss = reinforce_loss + bl_loss

    # Perform backward pass and optimization step
    optimizer.zero_grad()
    loss.backward()
    # Clip gradient norms and get (clipped) gradient norms for logging
    grad_norms = clip_grad_norms(optimizer.param_groups, opts.max_grad_norm)
    optimizer.step()

    # Logging
    if step % int(opts.log_step) == 0:
        log_values(cost, grad_norms, epoch, batch_id, step,
                   log_likelihood, reinforce_loss, bl_loss, tb_logger, opts)