model_l1.py 15.2 KB
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429
from __future__ import print_function, division

# pytorch imports
import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets, models, transforms
from torchvision import transforms, utils
from tensorboardX import SummaryWriter

# general imports
import os
import time
from shutil import rmtree

# data science imports
import csv

import cxr_dataset as CXR
import eval_model as E

use_gpu = torch.cuda.is_available()
gpu_count = torch.cuda.device_count()
print("Available GPU count:" + str(gpu_count))
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")


def checkpoint(model, best_loss, epoch, LR, filename):
    """
    Saves checkpoint of torchvision model during training.

    Args:
        model: torchvision model to be saved
        best_loss: best val loss achieved so far in training
        epoch: current epoch of training
        LR: current learning rate in training
    Returns:
        None
    """

    print('saving')
    state = {
        'model': model,
        'best_loss': best_loss,
        'epoch': epoch,
        'rng_state': torch.get_rng_state(),
        'LR': LR
    }

    torch.save(state, 'results/' + filename)


def pos_neg_weights_in_batch(labels_batch):

    num_total = labels_batch.shape[0] * labels_batch.shape[1]
    num_positives = labels_batch.sum()
    num_negatives = num_total - num_positives

    if not num_positives == 0:
        beta_p = num_negatives / num_positives
    else:
        beta_p = num_negatives
    beta_p = torch.as_tensor(beta_p)
    beta_p = beta_p.to(device)
    beta_p = beta_p.type(torch.cuda.FloatTensor)

    return beta_p


def train_model(
        model,
        criterion,
        optimizer,
        LR,
        num_epochs,
        dataloaders,
        dataset_sizes,
        weight_decay,
        weighted_cross_entropy_batchwise=False,
        fine_tune=False,
        regression=False):
    """
    Fine tunes torchvision model to NIH CXR data.

    Args:
        model: torchvision model to be finetuned (densenet-121 in this case)
        criterion: loss criterion (binary cross entropy loss, BCELoss)
        optimizer: optimizer to use in training (SGD)
        LR: learning rate
        num_epochs: continue training up to this many epochs
        dataloaders: pytorch train and val dataloaders
        dataset_sizes: length of train and val datasets
        weight_decay: weight decay parameter we use in SGD with momentum
    Returns:
        model: trained torchvision model
        best_epoch: epoch on which best model val loss was obtained
    """
    since = time.time()

    start_epoch = 1
    best_loss = 999999
    best_epoch = -1
    last_train_loss = -1

    tensorboard_writer_train = SummaryWriter('runs/loss/train_loss')
    tensorboard_writer_val = SummaryWriter('runs/loss/val_loss')

    if not fine_tune:
        PRED_LABEL = [
            'Atelectasis',
            'Cardiomegaly',
            'Effusion',
            'Infiltration',
            'Mass',
            'Nodule',
            'Pneumonia',
            'Pneumothorax',
            'Consolidation',
            'Edema',
            'Emphysema',
            'Fibrosis',
            'Pleural_Thickening',
            'Hernia']
    else:
        PRED_LABEL = [
            'Detector01',
            'Detector2',
            'Detector3']

    if not regression:
        tensorboard_writer_auc = {}
        tensorboard_writer_AP = {}
        for label in PRED_LABEL:
            tensorboard_writer_auc[label] = SummaryWriter('runs/auc/'+label)
            tensorboard_writer_AP[label] = SummaryWriter('runs/ap/' + label)
    else:
        tensorboard_writer_mae = SummaryWriter('runs/mae')

    # iterate over epochs
    for epoch in range(start_epoch, num_epochs + 1):
        print('Epoch {}/{}'.format(epoch, num_epochs))
        print('-' * 10)

        # set model to train or eval mode based on whether we are in train or
        # val; necessary to get correct predictions given batchnorm
        for phase in ['train', 'val']:
            if phase == 'train':
                model.train(True)
            else:
                model.train(False)

            running_loss = 0.0

            total_done = 0

            for data in dataloaders[phase]:
                if not regression:
                    inputs, labels, _ = data
                else:
                    inputs, ground_truths, _ = data
                batch_size = inputs.shape[0]
                inputs = inputs.to(device)
                if not regression:
                    labels = (labels.to(device)).float()
                else:
                    ground_truths = (ground_truths.to(device)).float()

                with torch.set_grad_enabled(phase == 'train'):

                    outputs = model(inputs)

                    # calculate gradient and update parameters in train phase
                    optimizer.zero_grad()

                    if weighted_cross_entropy_batchwise:
                        beta = pos_neg_weights_in_batch(labels)
                        criterion = nn.BCEWithLogitsLoss(pos_weight=beta)

                    if not regression:
                        loss = criterion(outputs, labels)
                    else:
                        ground_truths = ground_truths.unsqueeze(1)
                        loss = criterion(outputs, ground_truths)

                    if phase == 'train':
                        loss.backward()
                        optimizer.step()

                    running_loss += loss.item() * batch_size

            epoch_loss = running_loss / dataset_sizes[phase]

            if phase == 'train':
                tensorboard_writer_train.add_scalar('Loss', epoch_loss, epoch)
                last_train_loss = epoch_loss
            elif phase == 'val':
                tensorboard_writer_val.add_scalar('Loss', epoch_loss, epoch)

                if not regression:
                    preds, aucs = E.make_pred_multilabel(dataloaders['val'], model, save_as_csv=False, fine_tune=fine_tune)
                    aucs.set_index('label', inplace=True)
                    print(aucs)
                    for label in PRED_LABEL:
                        tensorboard_writer_auc[label].add_scalar('AUC', aucs.loc[label, 'auc'], epoch)
                        tensorboard_writer_AP[label].add_scalar('AP', aucs.loc[label, 'AP'], epoch)
                else:
                    mae, _, _ = E.evaluate_mae(dataloaders['val'], model)
                    print('MAE: ', mae)
                    tensorboard_writer_mae.add_scalar('MAE', mae, epoch)

            print(phase + ' epoch {}:loss {:.4f} with data size {}'.format(
                epoch, epoch_loss, dataset_sizes[phase]))

            # checkpoint model if has best val loss yet
            if phase == 'val' and epoch_loss < best_loss:
                best_loss = epoch_loss
                best_epoch = epoch
                if not fine_tune:
                    checkpoint(model, best_loss, epoch, LR, filename='checkpoint_best_l1')
                elif fine_tune and not regression:
                    checkpoint(model, best_loss, epoch, LR, filename='classification_checkpoint_best')
                else:
                    checkpoint(model, best_loss, epoch, LR, filename='regression_checkpoint_best_l1')

        # log training and validation loss over each epoch
        with open("results/log_train", 'a') as logfile:
            logwriter = csv.writer(logfile, delimiter=',')
            if epoch == 1:
                logwriter.writerow(["epoch", "train_loss", "val_loss"])
            logwriter.writerow([epoch, last_train_loss, epoch_loss])

        # Save model after each epoch
        # checkpoint(model, best_loss, epoch, LR, filename='checkpoint')

        total_done += batch_size
        if total_done % (100 * batch_size) == 0:
            print("completed " + str(total_done) + " so far in epoch")

        # print elapsed time from the beginning after each epoch
        print('Training complete in {:.0f}m {:.0f}s'.format(
            (time.time() - since) // 60, (time.time() - since) % 60))

    # total time
    time_elapsed = time.time() - since
    print('Training complete in {:.0f}m {:.0f}s'.format(
        time_elapsed // 60, time_elapsed % 60))

    # load best model weights to return
    if not fine_tune:
        checkpoint_best = torch.load('results/checkpoint_best_l1')
    elif fine_tune and not regression:
        checkpoint_best = torch.load('results/classification_checkpoint_best')
    else:
        checkpoint_best = torch.load('results/regression_checkpoint_best_l1')
    model = checkpoint_best['model']
    return model, best_epoch


def train_cnn(PATH_TO_IMAGES, LR, WEIGHT_DECAY, fine_tune=False, regression=False, freeze=False, adam=False,
              initial_model_path=None, initial_brixia_model_path=None, weighted_cross_entropy_batchwise=False,
              modification=None, weighted_cross_entropy=False):
    """
    Train torchvision model to NIH data given high level hyperparameters.

    Args:
        PATH_TO_IMAGES: path to NIH images
        LR: learning rate
        WEIGHT_DECAY: weight decay parameter for SGD

    Returns:
        preds: torchvision model predictions on test fold with ground truth for comparison
        aucs: AUCs for each train,test tuple

    """
    NUM_EPOCHS = 100
    BATCH_SIZE = 32

    try:
        rmtree('results/')
    except BaseException:
        pass  # directory doesn't yet exist, no need to clear it
    os.makedirs("results/")

    # use imagenet mean,std for normalization
    mean = [0.485, 0.456, 0.406]
    std = [0.229, 0.224, 0.225]

    N_LABELS = 14  # we are predicting 14 labels
    N_COVID_LABELS = 3  # we are predicting 3 COVID labels

    # define torchvision transforms
    data_transforms = {
        'train': transforms.Compose([
            # transforms.RandomHorizontalFlip(),
            transforms.Resize(224),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize(mean, std)
        ]),
        'val': transforms.Compose([
            transforms.Resize(224),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize(mean, std)
        ]),
    }

    # create train/val dataloaders
    transformed_datasets = {}
    transformed_datasets['train'] = CXR.CXRDataset(
        path_to_images=PATH_TO_IMAGES,
        fold='train',
        transform=data_transforms['train'],
        fine_tune=fine_tune,
        regression=regression)
    transformed_datasets['val'] = CXR.CXRDataset(
        path_to_images=PATH_TO_IMAGES,
        fold='val',
        transform=data_transforms['val'],
        fine_tune=fine_tune,
        regression=regression)

    dataloaders = {}
    dataloaders['train'] = torch.utils.data.DataLoader(
        transformed_datasets['train'],
        batch_size=BATCH_SIZE,
        shuffle=True,
        num_workers=8)
    dataloaders['val'] = torch.utils.data.DataLoader(
        transformed_datasets['val'],
        batch_size=BATCH_SIZE,
        shuffle=True,
        num_workers=8)

    # please do not attempt to train without GPU as will take excessively long
    if not use_gpu:
        raise ValueError("Error, requires GPU")

    if initial_model_path or initial_brixia_model_path:
        if initial_model_path:
            saved_model = torch.load(initial_model_path)
        else:
            saved_model = torch.load(initial_brixia_model_path)
        model = saved_model['model']
        del saved_model
        if fine_tune and not initial_brixia_model_path:
            num_ftrs = model.module.classifier.in_features
            if freeze:
                for feature in model.module.features:
                    for param in feature.parameters():
                        param.requires_grad = False
                    if feature == model.module.features.transition2:
                        break
            if not regression:
                model.module.classifier = nn.Linear(num_ftrs, N_COVID_LABELS)
            else:
                model.module.classifier = nn.Sequential(
                    nn.Linear(num_ftrs, 1),
                    nn.ReLU(inplace=True)
                )
    else:
        model = models.densenet121(pretrained=True)
        num_ftrs = model.classifier.in_features
        model.classifier = nn.Linear(num_ftrs, N_LABELS)

        if modification == 'transition_layer':
            # num_ftrs = model.features.norm5.num_features
            up1 = torch.nn.Sequential(torch.nn.ConvTranspose2d(num_ftrs, num_ftrs, kernel_size=3, stride=2, padding=1),
                                      torch.nn.BatchNorm2d(num_ftrs),
                                      torch.nn.ReLU(True))
            up2 = torch.nn.Sequential(torch.nn.ConvTranspose2d(num_ftrs, num_ftrs, kernel_size=3, stride=2, padding=1),
                                      torch.nn.BatchNorm2d(num_ftrs))

            transition_layer = torch.nn.Sequential(up1, up2)
            model.features.add_module('transition_chestX', transition_layer)

        if modification == 'remove_last_block':
            model.features.denseblock4 = nn.Sequential()
            model.features.transition3 = nn.Sequential()
            # model.features.norm5 = nn.BatchNorm2d(512)
            # model.classifier = nn.Linear(512, N_LABELS)
        if modification == 'remove_last_two_block':
            model.features.denseblock4 = nn.Sequential()
            model.features.transition3 = nn.Sequential()

            model.features.transition2 = nn.Sequential()
            model.features.denseblock3 = nn.Sequential()

            model.features.norm5 = nn.BatchNorm2d(512)
            model.classifier = nn.Linear(512, N_LABELS)

    print(model)

    # put model on GPU
    if not initial_model_path:
        model = nn.DataParallel(model)
    model.to(device)

    if regression:
        criterion = nn.L1Loss()
    else:
        if weighted_cross_entropy:
            pos_weights = transformed_datasets['train'].pos_neg_balance_weights()
            print(pos_weights)
            # pos_weights[pos_weights>40] = 40
            criterion = nn.BCEWithLogitsLoss(pos_weight=pos_weights)
        else:
            criterion = nn.BCEWithLogitsLoss()

    if adam:
        optimizer = optim.Adam(filter(lambda p: p.requires_grad, model.parameters()), lr=LR, weight_decay=WEIGHT_DECAY)
    else:
        optimizer = optim.SGD(filter(lambda p: p.requires_grad, model.parameters()), lr=LR, weight_decay=WEIGHT_DECAY, momentum=0.9)

    dataset_sizes = {x: len(transformed_datasets[x]) for x in ['train', 'val']}

    # train model
    if regression:
        model, best_epoch = train_model(model, criterion, optimizer, LR, num_epochs=NUM_EPOCHS,
                                        dataloaders=dataloaders, dataset_sizes=dataset_sizes,
                                        weight_decay=WEIGHT_DECAY, fine_tune=fine_tune, regression=regression)
    else:
        model, best_epoch = train_model(model, criterion, optimizer, LR, num_epochs=NUM_EPOCHS,
                                        dataloaders=dataloaders, dataset_sizes=dataset_sizes, weight_decay=WEIGHT_DECAY,
                                        weighted_cross_entropy_batchwise=weighted_cross_entropy_batchwise,
                                        fine_tune=fine_tune)
        # get preds and AUCs on test fold
        preds, aucs = E.make_pred_multilabel(dataloaders['val'], model, save_as_csv=False, fine_tune=fine_tune)
        return preds, aucs