def main(argv): fmt = '[%(filename)s:%(lineno)s] %(message)s' formatter = logging.PythonFormatter(fmt) logging.get_absl_handler().setFormatter(formatter) del argv # unused arg tf.io.gfile.makedirs(FLAGS.output_dir) logging.info('Saving checkpoints at %s', FLAGS.output_dir) tf.random.set_seed(FLAGS.seed) data_dir = FLAGS.data_dir if FLAGS.use_gpu: logging.info('Use GPU') strategy = tf.distribute.MirroredStrategy() else: logging.info('Use TPU at %s', FLAGS.tpu if FLAGS.tpu is not None else 'local') resolver = tf.distribute.cluster_resolver.TPUClusterResolver(tpu=FLAGS.tpu) tf.config.experimental_connect_to_cluster(resolver) tf.tpu.experimental.initialize_tpu_system(resolver) strategy = tf.distribute.TPUStrategy(resolver) ds_info = tfds.builder(FLAGS.dataset).info batch_size = FLAGS.per_core_batch_size * FLAGS.num_cores train_dataset_size = ( ds_info.splits['train'].num_examples * FLAGS.train_proportion) steps_per_epoch = int(train_dataset_size / batch_size) logging.info('Steps per epoch %s', steps_per_epoch) logging.info('Size of the dataset %s', ds_info.splits['train'].num_examples) logging.info('Train proportion %s', FLAGS.train_proportion) steps_per_eval = ds_info.splits['test'].num_examples // batch_size num_classes = ds_info.features['label'].num_classes aug_params = { 'augmix': FLAGS.augmix, 'aug_count': FLAGS.aug_count, 'augmix_depth': FLAGS.augmix_depth, 'augmix_prob_coeff': FLAGS.augmix_prob_coeff, 'augmix_width': FLAGS.augmix_width, } # Note that stateless_{fold_in,split} may incur a performance cost, but a # quick side-by-side test seemed to imply this was minimal. seeds = tf.random.experimental.stateless_split( [FLAGS.seed, FLAGS.seed + 1], 2)[:, 0] train_builder = ub.datasets.get( FLAGS.dataset, data_dir=data_dir, download_data=FLAGS.download_data, split=tfds.Split.TRAIN, seed=seeds[0], aug_params=aug_params, shuffle_buffer_size=FLAGS.shuffle_buffer_size, validation_percent=1. - FLAGS.train_proportion, ) train_dataset = train_builder.load(batch_size=batch_size) validation_dataset = None steps_per_validation = 0 if FLAGS.train_proportion < 1.0: validation_builder = ub.datasets.get( FLAGS.dataset, split=tfds.Split.VALIDATION, validation_percent=1. - FLAGS.train_proportion, data_dir=data_dir, drop_remainder=FLAGS.drop_remainder_for_eval) validation_dataset = validation_builder.load(batch_size=batch_size) validation_dataset = strategy.experimental_distribute_dataset( validation_dataset) steps_per_validation = validation_builder.num_examples // batch_size clean_test_builder = ub.datasets.get( FLAGS.dataset, split=tfds.Split.TEST, data_dir=data_dir, drop_remainder=FLAGS.drop_remainder_for_eval) clean_test_dataset = clean_test_builder.load(batch_size=batch_size) test_datasets = { 'clean': strategy.experimental_distribute_dataset(clean_test_dataset), } train_dataset = strategy.experimental_distribute_dataset(train_dataset) steps_per_epoch = train_builder.num_examples // batch_size steps_per_eval = clean_test_builder.num_examples // batch_size num_classes = 100 if FLAGS.dataset == 'cifar100' else 10 if FLAGS.eval_on_ood: ood_dataset_names = FLAGS.ood_dataset ood_ds, steps_per_ood = ood_utils.load_ood_datasets( ood_dataset_names, clean_test_builder, 1. - FLAGS.train_proportion, batch_size, drop_remainder=FLAGS.drop_remainder_for_eval) ood_datasets = { name: strategy.experimental_distribute_dataset(ds) for name, ds in ood_ds.items() } if FLAGS.corruptions_interval > 0: if FLAGS.dataset == 'cifar100': data_dir = FLAGS.cifar100_c_path corruption_types, _ = utils.load_corrupted_test_info(FLAGS.dataset) for corruption_type in corruption_types: for severity in range(1, 6): dataset = ub.datasets.get( f'{FLAGS.dataset}_corrupted', corruption_type=corruption_type, severity=severity, split=tfds.Split.TEST, data_dir=data_dir).load(batch_size=batch_size) test_datasets[f'{corruption_type}_{severity}'] = ( strategy.experimental_distribute_dataset(dataset)) summary_writer = tf.summary.create_file_writer( os.path.join(FLAGS.output_dir, 'summaries')) with strategy.scope(): logging.info('Building ResNet model') model = ub.models.wide_resnet( input_shape=(32, 32, 3), depth=28, width_multiplier=10, num_classes=num_classes, l2=FLAGS.l2, hps=_extract_hyperparameter_dictionary(), seed=seeds[1]) logging.info('Model input shape: %s', model.input_shape) logging.info('Model output shape: %s', model.output_shape) logging.info('Model number of weights: %s', model.count_params()) # Linearly scale learning rate and the decay epochs by vanilla settings. base_lr = FLAGS.base_learning_rate * batch_size / 128 lr_decay_epochs = [(int(start_epoch_str) * FLAGS.train_epochs) // 200 for start_epoch_str in FLAGS.lr_decay_epochs] lr_schedule = ub.schedules.WarmUpPiecewiseConstantSchedule( steps_per_epoch, base_lr, decay_ratio=FLAGS.lr_decay_ratio, decay_epochs=lr_decay_epochs, warmup_epochs=FLAGS.lr_warmup_epochs) optimizer = tf.keras.optimizers.SGD(lr_schedule, momentum=1.0 - FLAGS.one_minus_momentum, nesterov=True) metrics = { 'train/negative_log_likelihood': tf.keras.metrics.Mean(), 'train/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'train/loss': tf.keras.metrics.Mean(), 'train/ece': rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins), 'test/negative_log_likelihood': tf.keras.metrics.Mean(), 'test/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'test/ece': rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins), } if validation_dataset: metrics.update({ 'validation/negative_log_likelihood': tf.keras.metrics.Mean(), 'validation/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'validation/ece': rm.metrics.ExpectedCalibrationError( num_bins=FLAGS.num_bins), }) if FLAGS.eval_on_ood: ood_metrics = ood_utils.create_ood_metrics(ood_dataset_names) metrics.update(ood_metrics) if FLAGS.corruptions_interval > 0: corrupt_metrics = {} for intensity in range(1, 6): for corruption in corruption_types: dataset_name = '{0}_{1}'.format(corruption, intensity) corrupt_metrics['test/nll_{}'.format(dataset_name)] = ( tf.keras.metrics.Mean()) corrupt_metrics['test/accuracy_{}'.format(dataset_name)] = ( tf.keras.metrics.SparseCategoricalAccuracy()) corrupt_metrics['test/ece_{}'.format(dataset_name)] = ( rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins)) checkpoint = tf.train.Checkpoint(model=model, optimizer=optimizer) latest_checkpoint = tf.train.latest_checkpoint(FLAGS.output_dir) initial_epoch = 0 if latest_checkpoint: # checkpoint.restore must be within a strategy.scope() so that optimizer # slot variables are mirrored. checkpoint.restore(latest_checkpoint) logging.info('Loaded checkpoint %s', latest_checkpoint) initial_epoch = optimizer.iterations.numpy() // steps_per_epoch if FLAGS.saved_model_dir: logging.info('Saved model dir : %s', FLAGS.saved_model_dir) latest_checkpoint = tf.train.latest_checkpoint(FLAGS.saved_model_dir) checkpoint.restore(latest_checkpoint) logging.info('Loaded checkpoint %s', latest_checkpoint) if FLAGS.eval_only: initial_epoch = FLAGS.train_epochs - 1 # Run just one epoch of eval @tf.function def train_step(iterator): """Training StepFn.""" def step_fn(inputs): """Per-Replica StepFn.""" images = inputs['features'] labels = inputs['labels'] if FLAGS.augmix and FLAGS.aug_count >= 1: # Index 0 at augmix processing is the unperturbed image. # We take just 1 augmented image from the returned augmented images. images = images[:, 1, ...] with tf.GradientTape() as tape: logits = model(images, training=True) if FLAGS.label_smoothing == 0.: negative_log_likelihood = tf.reduce_mean( tf.keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)) else: one_hot_labels = tf.one_hot(tf.cast(labels, tf.int32), num_classes) negative_log_likelihood = tf.reduce_mean( tf.keras.losses.categorical_crossentropy( one_hot_labels, logits, from_logits=True, label_smoothing=FLAGS.label_smoothing)) l2_loss = sum(model.losses) loss = negative_log_likelihood + l2_loss # Scale the loss given the TPUStrategy will reduce sum all gradients. scaled_loss = loss / strategy.num_replicas_in_sync grads = tape.gradient(scaled_loss, model.trainable_variables) optimizer.apply_gradients(zip(grads, model.trainable_variables)) probs = tf.nn.softmax(logits) metrics['train/ece'].add_batch(probs, label=labels) metrics['train/loss'].update_state(loss) metrics['train/negative_log_likelihood'].update_state( negative_log_likelihood) metrics['train/accuracy'].update_state(labels, logits) for _ in tf.range(tf.cast(steps_per_epoch, tf.int32)): strategy.run(step_fn, args=(next(iterator),)) @tf.function def test_step(iterator, dataset_split, dataset_name, num_steps): """Evaluation StepFn.""" def step_fn(inputs): """Per-Replica StepFn.""" images = inputs['features'] labels = inputs['labels'] logits = model(images, training=False) probs = tf.nn.softmax(logits) negative_log_likelihood = tf.reduce_mean( tf.keras.losses.sparse_categorical_crossentropy(labels, probs)) if dataset_name == 'clean': metrics[f'{dataset_split}/negative_log_likelihood'].update_state( negative_log_likelihood) metrics[f'{dataset_split}/accuracy'].update_state(labels, probs) metrics[f'{dataset_split}/ece'].add_batch(probs, label=labels) elif dataset_name.startswith('ood/'): ood_labels = 1 - inputs['is_in_distribution'] if FLAGS.dempster_shafer_ood: ood_scores = ood_utils.DempsterShaferUncertainty(logits) else: ood_scores = 1 - tf.reduce_max(probs, axis=-1) # Edgecase for if dataset_name contains underscores for name, metric in metrics.items(): if dataset_name in name: metric.update_state(ood_labels, ood_scores) else: corrupt_metrics['test/nll_{}'.format(dataset_name)].update_state( negative_log_likelihood) corrupt_metrics['test/accuracy_{}'.format(dataset_name)].update_state( labels, probs) corrupt_metrics['test/ece_{}'.format(dataset_name)].add_batch( probs, label=labels) for _ in tf.range(tf.cast(num_steps, tf.int32)): strategy.run(step_fn, args=(next(iterator),)) @tf.function def cifar10h_test_step(iterator, num_steps): """Evaluation StepFn.""" def step_fn(inputs): """Per-Replica StepFn.""" images = inputs['features'] labels = inputs['labels'] logits = model(images, training=False) negative_log_likelihood = tf.keras.losses.CategoricalCrossentropy( from_logits=True, reduction=tf.keras.losses.Reduction.NONE)(labels, logits) negative_log_likelihood = tf.reduce_mean(negative_log_likelihood) metrics['cifar10h/nll'].update_state(negative_log_likelihood) label_diversity, sample_diversity, ged = _generalized_energy_distance( labels, tf.nn.softmax(logits), 10) metrics['cifar10h/ged'].update_state(ged) metrics['cifar10h/ged_label_diversity'].update_state( tf.reduce_mean(label_diversity)) metrics['cifar10h/ged_sample_diversity'].update_state( tf.reduce_mean(sample_diversity)) for _ in tf.range(tf.cast(num_steps, tf.int32)): strategy.run(step_fn, args=(next(iterator),)) metrics.update({'test/ms_per_example': tf.keras.metrics.Mean()}) metrics.update({'train/ms_per_example': tf.keras.metrics.Mean()}) train_iterator = iter(train_dataset) start_time = time.time() tb_callback = None if FLAGS.collect_profile: tb_callback = tf.keras.callbacks.TensorBoard( profile_batch=(100, 102), log_dir=os.path.join(FLAGS.output_dir, 'logs')) tb_callback.set_model(model) for epoch in range(initial_epoch, FLAGS.train_epochs): logging.info('Starting to run epoch: %s', epoch) if tb_callback: tb_callback.on_epoch_begin(epoch) if not FLAGS.eval_only: train_start_time = time.time() train_step(train_iterator) ms_per_example = (time.time() - train_start_time) * 1e6 / batch_size metrics['train/ms_per_example'].update_state(ms_per_example) current_step = (epoch + 1) * steps_per_epoch max_steps = steps_per_epoch * FLAGS.train_epochs time_elapsed = time.time() - start_time steps_per_sec = float(current_step) / time_elapsed eta_seconds = (max_steps - current_step) / steps_per_sec message = ('{:.1%} completion: epoch {:d}/{:d}. {:.1f} steps/s. ' 'ETA: {:.0f} min. Time elapsed: {:.0f} min'.format( current_step / max_steps, epoch + 1, FLAGS.train_epochs, steps_per_sec, eta_seconds / 60, time_elapsed / 60)) logging.info(message) if tb_callback: tb_callback.on_epoch_end(epoch) if validation_dataset: validation_iterator = iter(validation_dataset) test_step( validation_iterator, 'validation', 'clean', steps_per_validation) datasets_to_evaluate = {'clean': test_datasets['clean']} if (FLAGS.corruptions_interval > 0 and (epoch + 1) % FLAGS.corruptions_interval == 0): datasets_to_evaluate = test_datasets for dataset_name, test_dataset in datasets_to_evaluate.items(): test_iterator = iter(test_dataset) logging.info('Testing on dataset %s', dataset_name) logging.info('Starting to run eval at epoch: %s', epoch) test_start_time = time.time() test_step(test_iterator, 'test', dataset_name, steps_per_eval) ms_per_example = (time.time() - test_start_time) * 1e6 / batch_size metrics['test/ms_per_example'].update_state(ms_per_example) logging.info('Done with testing on %s', dataset_name) if FLAGS.eval_on_ood: for ood_dataset_name, ood_dataset in ood_datasets.items(): ood_iterator = iter(ood_dataset) logging.info('Calculating OOD on dataset %s', ood_dataset_name) logging.info('Running OOD eval at epoch: %s', epoch) test_step(ood_iterator, 'test', ood_dataset_name, steps_per_ood[ood_dataset_name]) logging.info('Done with OOD eval on %s', ood_dataset_name) corrupt_results = {} if (FLAGS.corruptions_interval > 0 and (epoch + 1) % FLAGS.corruptions_interval == 0): corrupt_results = utils.aggregate_corrupt_metrics(corrupt_metrics, corruption_types) logging.info('Train Loss: %.4f, Accuracy: %.2f%%', metrics['train/loss'].result(), metrics['train/accuracy'].result() * 100) logging.info('Test NLL: %.4f, Accuracy: %.2f%%', metrics['test/negative_log_likelihood'].result(), metrics['test/accuracy'].result() * 100) total_results = {name: metric.result() for name, metric in metrics.items()} total_results.update(corrupt_results) # Metrics from Robustness Metrics (like ECE) will return a dict with a # single key/value, instead of a scalar. total_results = { k: (list(v.values())[0] if isinstance(v, dict) else v) for k, v in total_results.items() } with summary_writer.as_default(): for name, result in total_results.items(): tf.summary.scalar(name, result, step=epoch + 1) for metric in metrics.values(): metric.reset_states() if FLAGS.corruptions_interval > 0: for metric in corrupt_metrics.values(): metric.reset_states() if (FLAGS.checkpoint_interval > 0 and (epoch + 1) % FLAGS.checkpoint_interval == 0): checkpoint_name = checkpoint.save( os.path.join(FLAGS.output_dir, 'checkpoint')) logging.info('Saved checkpoint to %s', checkpoint_name) final_checkpoint_name = checkpoint.save( os.path.join(FLAGS.output_dir, 'checkpoint')) logging.info('Saved last checkpoint to %s', final_checkpoint_name) with summary_writer.as_default(): hp.hparams({ 'base_learning_rate': FLAGS.base_learning_rate, 'one_minus_momentum': FLAGS.one_minus_momentum, 'l2': FLAGS.l2, })
def main(config, output_dir): seed = config.get('seed', 0) rng = jax.random.PRNGKey(seed) tf.random.set_seed(seed) if config.get('data_dir'): logging.info('data_dir=%s', config.data_dir) logging.info('Output dir: %s', output_dir) save_checkpoint_path = None if config.get('checkpoint_steps'): gfile.makedirs(output_dir) save_checkpoint_path = os.path.join(output_dir, 'checkpoint.npz') # Create an asynchronous multi-metric writer. writer = metric_writers.create_default_writer( output_dir, just_logging=jax.process_index() > 0) # The pool is used to perform misc operations such as logging in async way. pool = multiprocessing.pool.ThreadPool() def write_note(note): if jax.process_index() == 0: logging.info('NOTE: %s', note) write_note('Initializing...') # Verify settings to make sure no checkpoints are accidentally missed. if config.get('keep_checkpoint_steps'): assert config.get('checkpoint_steps'), 'Specify `checkpoint_steps`.' assert config.keep_checkpoint_steps % config.checkpoint_steps == 0, ( f'`keep_checkpoint_steps` ({config.checkpoint_steps}) should be' f'divisible by `checkpoint_steps ({config.checkpoint_steps}).`') batch_size = config.batch_size batch_size_eval = config.get('batch_size_eval', batch_size) if (batch_size % jax.device_count() != 0 or batch_size_eval % jax.device_count() != 0): raise ValueError( f'Batch sizes ({batch_size} and {batch_size_eval}) must ' f'be divisible by device number ({jax.device_count()})') local_batch_size = batch_size // jax.process_count() local_batch_size_eval = batch_size_eval // jax.process_count() logging.info( 'Global batch size %d on %d hosts results in %d local batch size. ' 'With %d devices per host (%d devices total), that\'s a %d per-device ' 'batch size.', batch_size, jax.process_count(), local_batch_size, jax.local_device_count(), jax.device_count(), local_batch_size // jax.local_device_count()) write_note('Initializing train dataset...') rng, train_ds_rng = jax.random.split(rng) train_ds_rng = jax.random.fold_in(train_ds_rng, jax.process_index()) train_ds = input_utils.get_data( dataset=config.dataset, split=config.train_split, rng=train_ds_rng, process_batch_size=local_batch_size, preprocess_fn=preprocess_spec.parse( spec=config.pp_train, available_ops=preprocess_utils.all_ops()), shuffle_buffer_size=config.shuffle_buffer_size, prefetch_size=config.get('prefetch_to_host', 2), data_dir=config.get('data_dir')) # Start prefetching already. train_iter = input_utils.start_input_pipeline( train_ds, config.get('prefetch_to_device', 1)) write_note('Initializing val dataset(s)...') def _get_val_split(dataset, split, pp_eval, data_dir=None): # We do ceil rounding such that we include the last incomplete batch. nval_img = input_utils.get_num_examples( dataset, split=split, process_batch_size=local_batch_size_eval, drop_remainder=False, data_dir=data_dir) val_steps = int(np.ceil(nval_img / batch_size_eval)) logging.info('Running validation for %d steps for %s, %s', val_steps, dataset, split) if isinstance(pp_eval, str): pp_eval = preprocess_spec.parse( spec=pp_eval, available_ops=preprocess_utils.all_ops()) val_ds = input_utils.get_data(dataset=dataset, split=split, rng=None, process_batch_size=local_batch_size_eval, preprocess_fn=pp_eval, cache=config.get('val_cache', 'batched'), repeat_after_batching=True, shuffle=False, prefetch_size=config.get( 'prefetch_to_host', 2), drop_remainder=False, data_dir=data_dir) val_iter = input_utils.start_input_pipeline( val_ds, config.get('prefetch_to_device', 1)) return (val_iter, val_steps) val_iter_splits = { 'val': _get_val_split(config.dataset, split=config.val_split, pp_eval=config.pp_eval, data_dir=config.get('data_dir')) } if config.get('eval_on_cifar_10h'): cifar10_to_cifar10h_fn = data_uncertainty_utils.create_cifar10_to_cifar10h_fn( config.get('data_dir', None)) preprocess_fn = preprocess_spec.parse( spec=config.pp_eval_cifar_10h, available_ops=preprocess_utils.all_ops()) pp_eval = lambda ex: preprocess_fn(cifar10_to_cifar10h_fn(ex)) val_iter_splits['cifar_10h'] = _get_val_split( 'cifar10', split=config.get('cifar_10h_split') or 'test', pp_eval=pp_eval, data_dir=config.get('data_dir')) elif config.get('eval_on_imagenet_real'): imagenet_to_real_fn = data_uncertainty_utils.create_imagenet_to_real_fn( ) preprocess_fn = preprocess_spec.parse( spec=config.pp_eval_imagenet_real, available_ops=preprocess_utils.all_ops()) pp_eval = lambda ex: preprocess_fn(imagenet_to_real_fn(ex)) val_iter_imagenet_real, val_steps = _get_val_split( 'imagenet2012_real', split=config.get('imagenet_real_split') or 'validation', pp_eval=pp_eval, data_dir=config.get('data_dir')) val_iter_splits['imagenet_real'] = (val_iter_imagenet_real, val_steps) ood_ds = {} if config.get('ood_datasets') and config.get('ood_methods'): if config.get( 'ood_methods'): # config.ood_methods is not a empty list logging.info('loading OOD dataset = %s', config.get('ood_dataset')) ood_ds, ood_ds_names = ood_utils.load_ood_datasets( config.dataset, config.ood_datasets, config.ood_split, config.pp_eval, config.pp_eval_ood, config.ood_methods, config.train_split, config.get('data_dir'), _get_val_split, ) ntrain_img = input_utils.get_num_examples( config.dataset, split=config.train_split, process_batch_size=local_batch_size, data_dir=config.get('data_dir')) steps_per_epoch = int(ntrain_img / batch_size) if config.get('num_epochs'): total_steps = int(config.num_epochs * steps_per_epoch) assert not config.get( 'total_steps'), 'Set either num_epochs or total_steps' else: total_steps = config.total_steps logging.info('Total train data points: %d', ntrain_img) logging.info( 'Running for %d steps, that means %f epochs and %d steps per epoch', total_steps, total_steps * batch_size / ntrain_img, steps_per_epoch) write_note('Initializing model...') logging.info('config.model = %s', config.get('model')) model = ub.models.vision_transformer(num_classes=config.num_classes, **config.get('model', {})) # We want all parameters to be created in host RAM, not on any device, they'll # be sent there later as needed, otherwise we already encountered two # situations where we allocate them twice. @partial(jax.jit, backend='cpu') def init(rng): image_size = tuple(train_ds.element_spec['image'].shape[2:]) logging.info('image_size = %s', image_size) dummy_input = jnp.zeros((local_batch_size, ) + image_size, jnp.float32) params = flax.core.unfreeze(model.init(rng, dummy_input, train=False))['params'] # Set bias in the head to a low value, such that loss is small initially. params['head']['bias'] = jnp.full_like(params['head']['bias'], config.get('init_head_bias', 0)) # init head kernel to all zeros for fine-tuning if config.get('model_init'): params['head']['kernel'] = jnp.full_like(params['head']['kernel'], 0) return params rng, rng_init = jax.random.split(rng) params_cpu = init(rng_init) if jax.process_index() == 0: num_params = sum(p.size for p in jax.tree_flatten(params_cpu)[0]) parameter_overview.log_parameter_overview(params_cpu) writer.write_scalars(step=0, scalars={'num_params': num_params}) @partial(jax.pmap, axis_name='batch') def evaluation_fn(params, images, labels, mask): # Ignore the entries with all zero labels for evaluation. mask *= labels.max(axis=1) logits, out = model.apply({'params': flax.core.freeze(params)}, images, train=False) label_indices = config.get('label_indices') logging.info('!!! mask %s, label_indices %s', mask, label_indices) if label_indices: logits = logits[:, label_indices] # Note that logits and labels are usually of the shape [batch,num_classes]. # But for OOD data, when num_classes_ood > num_classes_ind, we need to # adjust labels to labels[:, :config.num_classes] to match the shape of # logits. That is just to avoid shape mismatch. The output losses does not # have any meaning for OOD data, because OOD not belong to any IND class. losses = getattr(train_utils, config.get('loss', 'sigmoid_xent'))( logits=logits, labels=labels[:, :( len(label_indices) if label_indices else config.num_classes)], reduction=False) loss = jax.lax.psum(losses * mask, axis_name='batch') top1_idx = jnp.argmax(logits, axis=1) # Extracts the label at the highest logit index for each image. top1_correct = jnp.take_along_axis(labels, top1_idx[:, None], axis=1)[:, 0] ncorrect = jax.lax.psum(top1_correct * mask, axis_name='batch') n = jax.lax.psum(mask, axis_name='batch') metric_args = jax.lax.all_gather( [logits, labels, out['pre_logits'], mask], axis_name='batch') return ncorrect, loss, n, metric_args @partial(jax.pmap, axis_name='batch') def cifar_10h_evaluation_fn(params, images, labels, mask): logits, out = model.apply({'params': flax.core.freeze(params)}, images, train=False) label_indices = config.get('label_indices') if label_indices: logits = logits[:, label_indices] losses = getattr(train_utils, config.get('loss', 'softmax_xent'))(logits=logits, labels=labels, reduction=False) loss = jax.lax.psum(losses, axis_name='batch') top1_idx = jnp.argmax(logits, axis=1) # Extracts the label at the highest logit index for each image. one_hot_labels = jnp.eye(10)[jnp.argmax(labels, axis=1)] top1_correct = jnp.take_along_axis(one_hot_labels, top1_idx[:, None], axis=1)[:, 0] ncorrect = jax.lax.psum(top1_correct, axis_name='batch') n = jax.lax.psum(one_hot_labels, axis_name='batch') metric_args = jax.lax.all_gather( [logits, labels, out['pre_logits'], mask], axis_name='batch') return ncorrect, loss, n, metric_args # Setup function for computing representation. @partial(jax.pmap, axis_name='batch') def representation_fn(params, images, labels, mask): _, outputs = model.apply({'params': flax.core.freeze(params)}, images, train=False) representation = outputs[config.fewshot.representation_layer] representation = jax.lax.all_gather(representation, 'batch') labels = jax.lax.all_gather(labels, 'batch') mask = jax.lax.all_gather(mask, 'batch') return representation, labels, mask # Load the optimizer from flax. opt_name = config.get('optim_name') write_note(f'Initializing {opt_name} optimizer...') opt_def = getattr(flax.optim, opt_name)(**config.get('optim', {})) # We jit this, such that the arrays that are created are created on the same # device as the input is, in this case the CPU. Else they'd be on device[0]. opt_cpu = jax.jit(opt_def.create)(params_cpu) weight_decay_rules = config.get('weight_decay', []) or [] rescale_value = config.lr.base if config.get( 'weight_decay_decouple') else 1. weight_decay_fn = train_utils.get_weight_decay_fn( weight_decay_rules=weight_decay_rules, rescale_value=rescale_value) @partial(jax.pmap, axis_name='batch', donate_argnums=(0, )) def update_fn(opt, lr, images, labels, rng): """Update step.""" measurements = {} # Get device-specific loss rng. rng, rng_model = jax.random.split(rng, 2) rng_model_local = jax.random.fold_in(rng_model, jax.lax.axis_index('batch')) def loss_fn(params, images, labels): logits, _ = model.apply({'params': flax.core.freeze(params)}, images, train=True, rngs={'dropout': rng_model_local}) label_indices = config.get('label_indices') if label_indices: logits = logits[:, label_indices] return getattr(train_utils, config.get('loss', 'sigmoid_xent'))(logits=logits, labels=labels) # Implementation considerations compared and summarized at # https://docs.google.com/document/d/1g3kMEvqu1DOawaflKNyUsIoQ4yIVEoyE5ZlIPkIl4Lc/edit?hl=en# l, g = train_utils.accumulate_gradient(jax.value_and_grad(loss_fn), opt.target, images, labels, config.get('grad_accum_steps')) l, g = jax.lax.pmean((l, g), axis_name='batch') # Log the gradient norm only if we need to compute it anyways (clipping) # or if we don't use grad_accum_steps, as they interact badly. if config.get('grad_accum_steps', 1) == 1 or config.get('grad_clip_norm'): grads, _ = jax.tree_flatten(g) l2_g = jnp.sqrt(sum([jnp.vdot(p, p) for p in grads])) measurements['l2_grads'] = l2_g # Optionally resize the global gradient to a maximum norm. We found this # useful in some cases across optimizers, hence it's in the main loop. if config.get('grad_clip_norm'): g_factor = jnp.minimum(1.0, config.grad_clip_norm / l2_g) g = jax.tree_util.tree_map(lambda p: g_factor * p, g) opt = opt.apply_gradient(g, learning_rate=lr) opt = opt.replace(target=weight_decay_fn(opt.target, lr)) params, _ = jax.tree_flatten(opt.target) measurements['l2_params'] = jnp.sqrt( sum([jnp.vdot(p, p) for p in params])) return opt, l, rng, measurements rng, train_loop_rngs = jax.random.split(rng) reint_params = ('head/kernel', 'head/bias') if config.get('only_eval', False) or not config.get('reint_head', True): reint_params = [] checkpoint_data = checkpoint_utils.maybe_load_checkpoint( train_loop_rngs=train_loop_rngs, save_checkpoint_path=save_checkpoint_path, init_optimizer=opt_cpu, init_params=params_cpu, init_fixed_model_states=None, default_reinit_params=reint_params, config=config, ) train_loop_rngs = checkpoint_data.train_loop_rngs opt_cpu = checkpoint_data.optimizer accumulated_train_time = checkpoint_data.accumulated_train_time write_note('Adapting the checkpoint model...') adapted_params = checkpoint_utils.adapt_upstream_architecture( init_params=params_cpu, loaded_params=opt_cpu.target) opt_cpu = opt_cpu.replace(target=adapted_params) write_note('Kicking off misc stuff...') first_step = int(opt_cpu.state.step) # Might be a DeviceArray type. if first_step == 0 and jax.process_index() == 0: writer.write_hparams(dict(config)) chrono = train_utils.Chrono(first_step, total_steps, batch_size, accumulated_train_time) # Note: switch to ProfileAllHosts() if you need to profile all hosts. # (Xprof data become much larger and take longer to load for analysis) profiler = periodic_actions.Profile( # Create profile after every restart to analyze pre-emption related # problems and assure we get similar performance in every run. logdir=output_dir, first_profile=first_step + 10) # Prepare the learning-rate and pre-fetch it to device to avoid delays. lr_fn = train_utils.create_learning_rate_schedule(total_steps, **config.get('lr', {})) # TODO(dusenberrymw): According to flax docs, prefetching shouldn't be # necessary for TPUs. lr_iter = train_utils.prefetch_scalar(map(lr_fn, range(total_steps)), config.get('prefetch_to_device', 1)) write_note(f'Replicating...\n{chrono.note}') opt_repl = flax_utils.replicate(opt_cpu) write_note(f'Initializing few-shotters...\n{chrono.note}') fewshotter = None if 'fewshot' in config and fewshot is not None: fewshotter = fewshot.FewShotEvaluator( representation_fn, config.fewshot, config.fewshot.get('batch_size') or batch_size_eval) checkpoint_writer = None # Note: we return the train loss, val loss, and fewshot best l2s for use in # reproducibility unit tests. train_loss = -jnp.inf val_loss = {val_name: -jnp.inf for val_name, _ in val_iter_splits.items()} fewshot_results = {'dummy': {(0, 1): -jnp.inf}} write_note(f'First step compilations...\n{chrono.note}') logging.info('first_step = %s', first_step) # Advance the iterators if we are restarting from an earlier checkpoint. # TODO(dusenberrymw): Look into checkpointing dataset state instead. if first_step > 0: write_note('Advancing iterators after resuming from a checkpoint...') lr_iter = itertools.islice(lr_iter, first_step, None) train_iter = itertools.islice(train_iter, first_step, None) # NOTE: Validation eval is only run on certain steps, so determine how many # times it was run previously. num_val_runs = sum( map( lambda i: train_utils.itstime(i, config.log_eval_steps, total_steps), range(1, first_step + 1))) for val_name, (val_iter, val_steps) in val_iter_splits.items(): val_iter = itertools.islice(val_iter, num_val_runs * val_steps, None) val_iter_splits[val_name] = (val_iter, val_steps) # Using a python integer for step here, because opt.state.step is allocated # on TPU during replication. for step, train_batch, lr_repl in zip( range(first_step + 1, total_steps + 1), train_iter, lr_iter): with jax.profiler.TraceAnnotation('train_step', step_num=step, _r=1): if not config.get('only_eval', False): opt_repl, loss_value, train_loop_rngs, extra_measurements = update_fn( opt_repl, lr_repl, train_batch['image'], train_batch['labels'], rng=train_loop_rngs) if jax.process_index() == 0: profiler(step) # Checkpoint saving if not config.get('only_eval', False) and train_utils.itstime( step, config.get('checkpoint_steps'), total_steps, process=0): write_note('Checkpointing...') chrono.pause() train_utils.checkpointing_timeout( checkpoint_writer, config.get('checkpoint_timeout', 1)) accumulated_train_time = chrono.accum_train_time # We need to transfer the weights over now or else we risk keeping them # alive while they'll be updated in a future step, creating hard to debug # memory errors (see b/160593526). Also, takes device 0's params only. opt_cpu = jax.tree_util.tree_map(lambda x: np.array(x[0]), opt_repl) # Check whether we want to keep a copy of the current checkpoint. copy_step = None if train_utils.itstime(step, config.get('keep_checkpoint_steps'), total_steps): write_note('Keeping a checkpoint copy...') copy_step = step # Checkpoint should be a nested dictionary or FLAX datataclasses from # `flax.struct`. Both can be present in a checkpoint. checkpoint_data = checkpoint_utils.CheckpointData( train_loop_rngs=train_loop_rngs, optimizer=opt_cpu, accumulated_train_time=accumulated_train_time) checkpoint_writer = pool.apply_async( checkpoint_utils.checkpoint_trained_model, (checkpoint_data, save_checkpoint_path, copy_step)) chrono.resume() # Report training progress if not config.get('only_eval', False) and train_utils.itstime( step, config.log_training_steps, total_steps, process=0): write_note('Reporting training progress...') train_loss = loss_value[ 0] # Keep to return for reproducibility tests. timing_measurements, note = chrono.tick(step) write_note(note) train_measurements = {} train_measurements.update({ 'learning_rate': lr_repl[0], 'training_loss': train_loss, }) train_measurements.update( flax.jax_utils.unreplicate(extra_measurements)) train_measurements.update(timing_measurements) writer.write_scalars(step, train_measurements) # Report validation performance if train_utils.itstime(step, config.log_eval_steps, total_steps): write_note('Evaluating on the validation set...') chrono.pause() for val_name, (val_iter, val_steps) in val_iter_splits.items(): # Sets up evaluation metrics. ece_num_bins = config.get('ece_num_bins', 15) auc_num_bins = config.get('auc_num_bins', 1000) ece = rm.metrics.ExpectedCalibrationError( num_bins=ece_num_bins) calib_auc = rm.metrics.CalibrationAUC( correct_pred_as_pos_label=False) oc_auc_0_5 = rm.metrics.OracleCollaborativeAUC( oracle_fraction=0.005, num_bins=auc_num_bins) oc_auc_1 = rm.metrics.OracleCollaborativeAUC( oracle_fraction=0.01, num_bins=auc_num_bins) oc_auc_2 = rm.metrics.OracleCollaborativeAUC( oracle_fraction=0.02, num_bins=auc_num_bins) oc_auc_5 = rm.metrics.OracleCollaborativeAUC( oracle_fraction=0.05, num_bins=auc_num_bins) label_diversity = tf.keras.metrics.Mean() sample_diversity = tf.keras.metrics.Mean() ged = tf.keras.metrics.Mean() # Runs evaluation loop. ncorrect, loss, nseen = 0, 0, 0 for _, batch in zip(range(val_steps), val_iter): if val_name == 'cifar_10h': batch_ncorrect, batch_losses, batch_n, batch_metric_args = ( cifar_10h_evaluation_fn(opt_repl.target, batch['image'], batch['labels'], batch['mask'])) else: batch_ncorrect, batch_losses, batch_n, batch_metric_args = ( evaluation_fn(opt_repl.target, batch['image'], batch['labels'], batch['mask'])) # All results are a replicated array shaped as follows: # (local_devices, per_device_batch_size, elem_shape...) # with each local device's entry being identical as they got psum'd. # So let's just take the first one to the host as numpy. ncorrect += np.sum(np.array(batch_ncorrect[0])) loss += np.sum(np.array(batch_losses[0])) nseen += np.sum(np.array(batch_n[0])) if config.get('loss', 'sigmoid_xent') != 'sigmoid_xent': # Here we parse batch_metric_args to compute uncertainty metrics. # (e.g., ECE or Calibration AUC). logits, labels, _, masks = batch_metric_args masks = np.array(masks[0], dtype=np.bool) logits = np.array(logits[0]) probs = jax.nn.softmax(logits) # From one-hot to integer labels, as required by ECE. int_labels = np.argmax(np.array(labels[0]), axis=-1) int_preds = np.argmax(logits, axis=-1) confidence = np.max(probs, axis=-1) for p, c, l, d, m, label in zip( probs, confidence, int_labels, int_preds, masks, labels[0]): ece.add_batch(p[m, :], label=l[m]) calib_auc.add_batch(d[m], label=l[m], confidence=c[m]) # TODO(jereliu): Extend to support soft multi-class probabilities. oc_auc_0_5.add_batch(d[m], label=l[m], custom_binning_score=c[m]) oc_auc_1.add_batch(d[m], label=l[m], custom_binning_score=c[m]) oc_auc_2.add_batch(d[m], label=l[m], custom_binning_score=c[m]) oc_auc_5.add_batch(d[m], label=l[m], custom_binning_score=c[m]) if val_name == 'cifar_10h' or val_name == 'imagenet_real': batch_label_diversity, batch_sample_diversity, batch_ged = data_uncertainty_utils.generalized_energy_distance( label[m], p[m, :], config.num_classes) label_diversity.update_state( batch_label_diversity) sample_diversity.update_state( batch_sample_diversity) ged.update_state(batch_ged) val_loss[ val_name] = loss / nseen # Keep for reproducibility tests. val_measurements = { f'{val_name}_prec@1': ncorrect / nseen, f'{val_name}_loss': val_loss[val_name], } if config.get('loss', 'sigmoid_xent') != 'sigmoid_xent': val_measurements[f'{val_name}_ece'] = ece.result()['ece'] val_measurements[ f'{val_name}_calib_auc'] = calib_auc.result( )['calibration_auc'] val_measurements[ f'{val_name}_oc_auc_0.5%'] = oc_auc_0_5.result( )['collaborative_auc'] val_measurements[ f'{val_name}_oc_auc_1%'] = oc_auc_1.result( )['collaborative_auc'] val_measurements[ f'{val_name}_oc_auc_2%'] = oc_auc_2.result( )['collaborative_auc'] val_measurements[ f'{val_name}_oc_auc_5%'] = oc_auc_5.result( )['collaborative_auc'] writer.write_scalars(step, val_measurements) if val_name == 'cifar_10h' or val_name == 'imagenet_real': cifar_10h_measurements = { f'{val_name}_label_diversity': label_diversity.result(), f'{val_name}_sample_diversity': sample_diversity.result(), f'{val_name}_ged': ged.result(), } writer.write_scalars(step, cifar_10h_measurements) # OOD eval # Entries in the ood_ds dict include: # (ind_dataset, ood_dataset1, ood_dataset2, ...). # OOD metrics are computed using ind_dataset paired with each of the # ood_dataset. When Mahalanobis distance method is applied, train_ind_ds # is also included in the ood_ds. if ood_ds and config.ood_methods: ood_measurements = ood_utils.eval_ood_metrics( ood_ds, ood_ds_names, config.ood_methods, evaluation_fn, opt_repl) writer.write_scalars(step, ood_measurements) chrono.resume() if 'fewshot' in config and fewshotter is not None: # Compute few-shot on-the-fly evaluation. if train_utils.itstime(step, config.fewshot.log_steps, total_steps): chrono.pause() write_note(f'Few-shot evaluation...\n{chrono.note}') # Keep `results` to return for reproducibility tests. fewshot_results, best_l2 = fewshotter.run_all( opt_repl.target, config.fewshot.datasets) # TODO(dusenberrymw): Remove this once fewshot.py is updated. def make_writer_measure_fn(step): def writer_measure(name, value): writer.write_scalars(step, {name: value}) return writer_measure fewshotter.walk_results(make_writer_measure_fn(step), fewshot_results, best_l2) chrono.resume() # End of step. if config.get('testing_failure_step'): # Break early to simulate infra failures in test cases. if config.testing_failure_step == step: break write_note(f'Done!\n{chrono.note}') pool.close() pool.join() writer.close() # Return final training loss, validation loss, and fewshot results for # reproducibility test cases. return train_loss, val_loss, fewshot_results
def main(config, output_dir): seed = config.get('seed', 0) tf.random.set_seed(seed) if config.get('data_dir'): logging.info('data_dir=%s', config.data_dir) logging.info('Output dir: %s', output_dir) tf.io.gfile.makedirs(output_dir) # Create an asynchronous multi-metric writer. writer = metric_writers.create_default_writer( output_dir, just_logging=jax.process_index() > 0) # The pool is used to perform misc operations such as logging in async way. pool = multiprocessing.pool.ThreadPool() def write_note(note): if jax.process_index() == 0: logging.info('NOTE: %s', note) write_note('Initializing...') batch_size = config.batch_size batch_size_eval = config.get('batch_size_eval', batch_size) if (batch_size % jax.device_count() != 0 or batch_size_eval % jax.device_count() != 0): raise ValueError( f'Batch sizes ({batch_size} and {batch_size_eval}) must ' f'be divisible by device number ({jax.device_count()})') local_batch_size = batch_size // jax.process_count() local_batch_size_eval = batch_size_eval // jax.process_count() logging.info( 'Global batch size %d on %d hosts results in %d local batch size. ' 'With %d devices per host (%d devices total), that\'s a %d per-device ' 'batch size.', batch_size, jax.process_count(), local_batch_size, jax.local_device_count(), jax.device_count(), local_batch_size // jax.local_device_count()) write_note('Initializing val dataset(s)...') def _get_val_split(dataset, split, pp_eval, data_dir=None): # We do ceil rounding such that we include the last incomplete batch. nval_img = input_utils.get_num_examples( dataset, split=split, process_batch_size=local_batch_size_eval, drop_remainder=False, data_dir=data_dir) val_steps = int(np.ceil(nval_img / batch_size_eval)) logging.info('Running validation for %d steps for %s, %s', val_steps, dataset, split) if isinstance(pp_eval, str): pp_eval = preprocess_spec.parse( spec=pp_eval, available_ops=preprocess_utils.all_ops()) val_ds = input_utils.get_data(dataset=dataset, split=split, rng=None, process_batch_size=local_batch_size_eval, preprocess_fn=pp_eval, cache=config.get('val_cache', 'batched'), num_epochs=1, repeat_after_batching=True, shuffle=False, prefetch_size=config.get( 'prefetch_to_host', 2), drop_remainder=False, data_dir=data_dir) return val_ds val_ds_splits = { 'val': _get_val_split(config.dataset, split=config.val_split, pp_eval=config.pp_eval, data_dir=config.get('data_dir')) } if config.get('test_split'): val_ds_splits.update({ 'test': _get_val_split(config.dataset, split=config.test_split, pp_eval=config.pp_eval, data_dir=config.get('data_dir')) }) if config.get('eval_on_cifar_10h'): cifar10_to_cifar10h_fn = data_uncertainty_utils.create_cifar10_to_cifar10h_fn( config.get('data_dir', None)) preprocess_fn = preprocess_spec.parse( spec=config.pp_eval_cifar_10h, available_ops=preprocess_utils.all_ops()) pp_eval = lambda ex: preprocess_fn(cifar10_to_cifar10h_fn(ex)) val_ds_splits['cifar_10h'] = _get_val_split( 'cifar10', split=config.get('cifar_10h_split') or 'test', pp_eval=pp_eval, data_dir=config.get('data_dir')) elif config.get('eval_on_imagenet_real'): imagenet_to_real_fn = data_uncertainty_utils.create_imagenet_to_real_fn( ) preprocess_fn = preprocess_spec.parse( spec=config.pp_eval_imagenet_real, available_ops=preprocess_utils.all_ops()) pp_eval = lambda ex: preprocess_fn(imagenet_to_real_fn(ex)) val_ds_splits['imagenet_real'] = _get_val_split( 'imagenet2012_real', split=config.get('imagenet_real_split') or 'validation', pp_eval=pp_eval, data_dir=config.get('data_dir')) ood_ds = {} if config.get('ood_datasets') and config.get('ood_methods'): if config.get( 'ood_methods'): # config.ood_methods is not a empty list logging.info('loading OOD dataset = %s', config.get('ood_datasets')) ood_ds, ood_ds_names = ood_utils.load_ood_datasets( config.dataset, config.ood_datasets, config.ood_split, config.pp_eval, config.pp_eval_ood, config.ood_methods, config.train_split, config.get('data_dir'), _get_val_split, ) write_note('Initializing model...') logging.info('config.model = %s', config.model) model = ub.models.vision_transformer(num_classes=config.num_classes, **config.model) ensemble_pred_fn = functools.partial(ensemble_prediction_fn, model.apply) @functools.partial(jax.pmap, axis_name='batch') def evaluation_fn(params, images, labels, mask): # params is a dict of the form: # {'model_1': params_model_1, 'model_2': params_model_2, ...} # Ignore the entries with all zero labels for evaluation. mask *= labels.max(axis=1) loss_as_str = config.get('loss', 'sigmoid_xent') ens_logits, ens_prelogits = ensemble_pred_fn(params, images, loss_as_str) label_indices = config.get('label_indices') logging.info('!!! mask %s, label_indices %s', mask, label_indices) if label_indices: ens_logits = ens_logits[:, label_indices] # Note that logits and labels are usually of the shape [batch,num_classes]. # But for OOD data, when num_classes_ood > num_classes_ind, we need to # adjust labels to labels[:, :config.num_classes] to match the shape of # logits. That is just to avoid shape mismatch. The output losses does not # have any meaning for OOD data, because OOD not belong to any IND class. losses = getattr(train_utils, loss_as_str)( logits=ens_logits, labels=labels[:, :( len(label_indices) if label_indices else config.num_classes)], reduction=False) loss = jax.lax.psum(losses * mask, axis_name='batch') top1_idx = jnp.argmax(ens_logits, axis=1) # Extracts the label at the highest logit index for each image. top1_correct = jnp.take_along_axis(labels, top1_idx[:, None], axis=1)[:, 0] ncorrect = jax.lax.psum(top1_correct * mask, axis_name='batch') n = jax.lax.psum(mask, axis_name='batch') metric_args = jax.lax.all_gather( [ens_logits, labels, ens_prelogits, mask], axis_name='batch') return ncorrect, loss, n, metric_args @functools.partial(jax.pmap, axis_name='batch') def cifar_10h_evaluation_fn(params, images, labels, mask): loss_as_str = config.get('loss', 'softmax_xent') ens_logits, ens_prelogits = ensemble_pred_fn(params, images, loss_as_str) label_indices = config.get('label_indices') if label_indices: ens_logits = ens_logits[:, label_indices] losses = getattr(train_utils, loss_as_str)(logits=ens_logits, labels=labels, reduction=False) loss = jax.lax.psum(losses, axis_name='batch') top1_idx = jnp.argmax(ens_logits, axis=1) # Extracts the label at the highest logit index for each image. one_hot_labels = jnp.eye(10)[jnp.argmax(labels, axis=1)] top1_correct = jnp.take_along_axis(one_hot_labels, top1_idx[:, None], axis=1)[:, 0] ncorrect = jax.lax.psum(top1_correct, axis_name='batch') n = jax.lax.psum(one_hot_labels, axis_name='batch') metric_args = jax.lax.all_gather( [ens_logits, labels, ens_prelogits, mask], axis_name='batch') return ncorrect, loss, n, metric_args # Setup function for computing representation. @functools.partial(jax.pmap, axis_name='batch') def representation_fn(params, images, labels, mask): # Return shape [batch_size, representation_size * ensemble_size]. During # few-shot eval, a single linear regressor is applied over all dimensions. representation = [] for p in params.values(): _, outputs = model.apply({'params': flax.core.freeze(p)}, images, train=False) representation += [outputs[config.fewshot.representation_layer]] representation = jnp.concatenate(representation, axis=1) representation = jax.lax.all_gather(representation, 'batch') labels = jax.lax.all_gather(labels, 'batch') mask = jax.lax.all_gather(mask, 'batch') return representation, labels, mask write_note('Load checkpoints...') ensemble_params = load_checkpoints(config) write_note('Replicating...') ensemble_params = flax.jax_utils.replicate(ensemble_params) if jax.process_index() == 0: writer.write_hparams(dict(config)) write_note('Initializing few-shotters...') fewshotter = None if 'fewshot' in config and fewshot is not None: fewshotter = fewshot.FewShotEvaluator( representation_fn, config.fewshot, config.fewshot.get('batch_size') or batch_size_eval) # Note: we return the train loss, val loss, and fewshot best l2s for use in # reproducibility unit tests. val_loss = {val_name: -jnp.inf for val_name, _ in val_ds_splits.items()} fewshot_results = {'dummy': {(0, 1): -jnp.inf}} step = 1 # Report validation performance. write_note('Evaluating on the validation set...') for val_name, val_ds in val_ds_splits.items(): # Sets up evaluation metrics. ece_num_bins = config.get('ece_num_bins', 15) auc_num_bins = config.get('auc_num_bins', 1000) ece = rm.metrics.ExpectedCalibrationError(num_bins=ece_num_bins) calib_auc = rm.metrics.CalibrationAUC(correct_pred_as_pos_label=False) oc_auc_0_5 = rm.metrics.OracleCollaborativeAUC(oracle_fraction=0.005, num_bins=auc_num_bins) oc_auc_1 = rm.metrics.OracleCollaborativeAUC(oracle_fraction=0.01, num_bins=auc_num_bins) oc_auc_2 = rm.metrics.OracleCollaborativeAUC(oracle_fraction=0.02, num_bins=auc_num_bins) oc_auc_5 = rm.metrics.OracleCollaborativeAUC(oracle_fraction=0.05, num_bins=auc_num_bins) label_diversity = tf.keras.metrics.Mean() sample_diversity = tf.keras.metrics.Mean() ged = tf.keras.metrics.Mean() # Runs evaluation loop. val_iter = input_utils.start_input_pipeline( val_ds, config.get('prefetch_to_device', 1)) ncorrect, loss, nseen = 0, 0, 0 for batch in val_iter: if val_name == 'cifar_10h': batch_ncorrect, batch_losses, batch_n, batch_metric_args = ( cifar_10h_evaluation_fn(ensemble_params, batch['image'], batch['labels'], batch['mask'])) else: batch_ncorrect, batch_losses, batch_n, batch_metric_args = ( evaluation_fn(ensemble_params, batch['image'], batch['labels'], batch['mask'])) # All results are a replicated array shaped as follows: # (local_devices, per_device_batch_size, elem_shape...) # with each local device's entry being identical as they got psum'd. # So let's just take the first one to the host as numpy. ncorrect += np.sum(np.array(batch_ncorrect[0])) loss += np.sum(np.array(batch_losses[0])) nseen += np.sum(np.array(batch_n[0])) if config.get('loss', 'sigmoid_xent') != 'sigmoid_xent': # Here we parse batch_metric_args to compute uncertainty metrics. # (e.g., ECE or Calibration AUC). logits, labels, _, masks = batch_metric_args masks = np.array(masks[0], dtype=np.bool) logits = np.array(logits[0]) probs = jax.nn.softmax(logits) # From one-hot to integer labels, as required by ECE. int_labels = np.argmax(np.array(labels[0]), axis=-1) int_preds = np.argmax(logits, axis=-1) confidence = np.max(probs, axis=-1) for p, c, l, d, m, label in zip(probs, confidence, int_labels, int_preds, masks, labels[0]): ece.add_batch(p[m, :], label=l[m]) calib_auc.add_batch(d[m], label=l[m], confidence=c[m]) # TODO(jereliu): Extend to support soft multi-class probabilities. oc_auc_0_5.add_batch(d[m], label=l[m], custom_binning_score=c[m]) oc_auc_1.add_batch(d[m], label=l[m], custom_binning_score=c[m]) oc_auc_2.add_batch(d[m], label=l[m], custom_binning_score=c[m]) oc_auc_5.add_batch(d[m], label=l[m], custom_binning_score=c[m]) if val_name == 'cifar_10h' or val_name == 'imagenet_real': batch_label_diversity, batch_sample_diversity, batch_ged = data_uncertainty_utils.generalized_energy_distance( label[m], p[m, :], config.num_classes) label_diversity.update_state(batch_label_diversity) sample_diversity.update_state(batch_sample_diversity) ged.update_state(batch_ged) val_loss[val_name] = loss / nseen # Keep for reproducibility tests. val_measurements = { f'{val_name}_prec@1': ncorrect / nseen, f'{val_name}_loss': val_loss[val_name], } if config.get('loss', 'sigmoid_xent') != 'sigmoid_xent': val_measurements[f'{val_name}_ece'] = ece.result()['ece'] val_measurements[f'{val_name}_calib_auc'] = calib_auc.result( )['calibration_auc'] val_measurements[f'{val_name}_oc_auc_0.5%'] = oc_auc_0_5.result( )['collaborative_auc'] val_measurements[f'{val_name}_oc_auc_1%'] = oc_auc_1.result( )['collaborative_auc'] val_measurements[f'{val_name}_oc_auc_2%'] = oc_auc_2.result( )['collaborative_auc'] val_measurements[f'{val_name}_oc_auc_5%'] = oc_auc_5.result( )['collaborative_auc'] writer.write_scalars(step, val_measurements) if val_name == 'cifar_10h' or val_name == 'imagenet_real': cifar_10h_measurements = { f'{val_name}_label_diversity': label_diversity.result(), f'{val_name}_sample_diversity': sample_diversity.result(), f'{val_name}_ged': ged.result(), } writer.write_scalars(step, cifar_10h_measurements) # OOD eval # Entries in the ood_ds dict include: # (ind_dataset, ood_dataset1, ood_dataset2, ...). # OOD metrics are computed using ind_dataset paired with each of the # ood_dataset. When Mahalanobis distance method is applied, train_ind_ds # is also included in the ood_ds. if ood_ds and config.ood_methods: ood_measurements = ood_utils.eval_ood_metrics(ood_ds, ood_ds_names, config.ood_methods, evaluation_fn, ensemble_params, n_prefetch=config.get( 'prefetch_to_device', 1)) writer.write_scalars(step, ood_measurements) if 'fewshot' in config and fewshotter is not None: # Compute few-shot on-the-fly evaluation. write_note('Few-shot evaluation...') # Keep `results` to return for reproducibility tests. fewshot_results, best_l2 = fewshotter.run_all(ensemble_params, config.fewshot.datasets) # TODO(dusenberrymw): Remove this once fewshot.py is updated. def make_writer_measure_fn(step): def writer_measure(name, value): writer.write_scalars(step, {name: value}) return writer_measure fewshotter.walk_results(make_writer_measure_fn(step), fewshot_results, best_l2) write_note('Done!') pool.close() pool.join() writer.close() # Return final training loss, validation loss, and fewshot results for # reproducibility test cases. return val_loss, fewshot_results
def main(argv): del argv # unused arg tf.io.gfile.makedirs(FLAGS.output_dir) logging.info('Saving checkpoints at %s', FLAGS.output_dir) tf.random.set_seed(FLAGS.seed) # Split the seed into a 2-tuple, for passing into dataset builder. dataset_seed = (FLAGS.seed, FLAGS.seed + 1) data_dir = FLAGS.data_dir if FLAGS.use_gpu: logging.info('Use GPU') strategy = tf.distribute.MirroredStrategy() else: logging.info('Use TPU at %s', FLAGS.tpu if FLAGS.tpu is not None else 'local') resolver = tf.distribute.cluster_resolver.TPUClusterResolver(tpu=FLAGS.tpu) tf.config.experimental_connect_to_cluster(resolver) tf.tpu.experimental.initialize_tpu_system(resolver) strategy = tf.distribute.TPUStrategy(resolver) batch_size = FLAGS.total_batch_size // FLAGS.num_dropout_samples_training test_batch_size = FLAGS.total_batch_size num_classes = 10 if FLAGS.dataset == 'cifar10' else 100 aug_params = { 'augmix': FLAGS.augmix, 'aug_count': FLAGS.aug_count, 'augmix_depth': FLAGS.augmix_depth, 'augmix_prob_coeff': FLAGS.augmix_prob_coeff, 'augmix_width': FLAGS.augmix_width, 'ensemble_size': 1, 'mixup_alpha': FLAGS.mixup_alpha, } validation_proportion = 1. - FLAGS.train_proportion use_validation_set = validation_proportion > 0. if FLAGS.dataset == 'cifar10': dataset_builder_class = ub.datasets.Cifar10Dataset else: dataset_builder_class = ub.datasets.Cifar100Dataset train_dataset_builder = dataset_builder_class( data_dir=data_dir, download_data=FLAGS.download_data, split=tfds.Split.TRAIN, use_bfloat16=FLAGS.use_bfloat16, aug_params=aug_params, validation_percent=validation_proportion, shuffle_buffer_size=FLAGS.shuffle_buffer_size, seed=dataset_seed) train_dataset = train_dataset_builder.load(batch_size=batch_size) if validation_proportion > 0.: validation_dataset_builder = dataset_builder_class( data_dir=data_dir, download_data=FLAGS.download_data, split=tfds.Split.VALIDATION, use_bfloat16=FLAGS.use_bfloat16, validation_percent=validation_proportion, drop_remainder=FLAGS.drop_remainder_for_eval) validation_dataset = validation_dataset_builder.load( batch_size=test_batch_size) validation_dataset = strategy.experimental_distribute_dataset( validation_dataset) val_sample_size = validation_dataset_builder.num_examples steps_per_val = steps_per_epoch = int(val_sample_size / test_batch_size) clean_test_dataset_builder = dataset_builder_class( data_dir=data_dir, download_data=FLAGS.download_data, split=tfds.Split.TEST, use_bfloat16=FLAGS.use_bfloat16, drop_remainder=FLAGS.drop_remainder_for_eval) clean_test_dataset = clean_test_dataset_builder.load( batch_size=test_batch_size) steps_per_epoch = train_dataset_builder.num_examples // batch_size steps_per_eval = clean_test_dataset_builder.num_examples // test_batch_size train_dataset = strategy.experimental_distribute_dataset(train_dataset) test_datasets = { 'clean': strategy.experimental_distribute_dataset(clean_test_dataset), } if FLAGS.eval_on_ood: ood_dataset_names = FLAGS.ood_dataset ood_ds, steps_per_ood = ood_utils.load_ood_datasets( ood_dataset_names, clean_test_dataset_builder, validation_proportion, test_batch_size, drop_remainder=FLAGS.drop_remainder_for_eval) ood_datasets = { name: strategy.experimental_distribute_dataset(ds) for name, ds in ood_ds.items() } if FLAGS.corruptions_interval > 0: if FLAGS.dataset == 'cifar100': data_dir = FLAGS.cifar100_c_path corruption_types, _ = utils.load_corrupted_test_info(FLAGS.dataset) for corruption_type in corruption_types: for severity in range(1, 6): dataset = ub.datasets.get( f'{FLAGS.dataset}_corrupted', corruption_type=corruption_type, severity=severity, split=tfds.Split.TEST, data_dir=data_dir, drop_remainder=FLAGS.drop_remainder_for_eval).load( batch_size=batch_size) test_datasets[f'{corruption_type}_{severity}'] = ( strategy.experimental_distribute_dataset(dataset)) if FLAGS.use_bfloat16: tf.keras.mixed_precision.set_global_policy('mixed_bfloat16') summary_writer = tf.summary.create_file_writer( os.path.join(FLAGS.output_dir, 'summaries')) with strategy.scope(): logging.info('Building ResNet model') if FLAGS.use_spec_norm: logging.info('Use Spectral Normalization with norm bound %.2f', FLAGS.spec_norm_bound) if FLAGS.use_gp_layer: logging.info('Use GP layer with hidden units %d', FLAGS.gp_hidden_dim) model = ub.models.wide_resnet_sngp( input_shape=(32, 32, 3), batch_size=batch_size, depth=28, width_multiplier=10, num_classes=num_classes, l2=FLAGS.l2, use_mc_dropout=FLAGS.use_mc_dropout, use_filterwise_dropout=FLAGS.use_filterwise_dropout, dropout_rate=FLAGS.dropout_rate, use_gp_layer=FLAGS.use_gp_layer, gp_input_dim=FLAGS.gp_input_dim, gp_hidden_dim=FLAGS.gp_hidden_dim, gp_scale=FLAGS.gp_scale, gp_bias=FLAGS.gp_bias, gp_input_normalization=FLAGS.gp_input_normalization, gp_random_feature_type=FLAGS.gp_random_feature_type, gp_cov_discount_factor=FLAGS.gp_cov_discount_factor, gp_cov_ridge_penalty=FLAGS.gp_cov_ridge_penalty, use_spec_norm=FLAGS.use_spec_norm, spec_norm_iteration=FLAGS.spec_norm_iteration, spec_norm_bound=FLAGS.spec_norm_bound) logging.info('Model input shape: %s', model.input_shape) logging.info('Model output shape: %s', model.output_shape) logging.info('Model number of weights: %s', model.count_params()) # Linearly scale learning rate and the decay epochs by vanilla settings. base_lr = FLAGS.base_learning_rate * batch_size / 128 lr_decay_epochs = [(int(start_epoch_str) * FLAGS.train_epochs) // 200 for start_epoch_str in FLAGS.lr_decay_epochs] lr_schedule = ub.schedules.WarmUpPiecewiseConstantSchedule( steps_per_epoch, base_lr, decay_ratio=FLAGS.lr_decay_ratio, decay_epochs=lr_decay_epochs, warmup_epochs=FLAGS.lr_warmup_epochs) optimizer = tf.keras.optimizers.SGD(lr_schedule, momentum=1.0 - FLAGS.one_minus_momentum, nesterov=True) metrics = { 'train/negative_log_likelihood': tf.keras.metrics.Mean(), 'train/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'train/loss': tf.keras.metrics.Mean(), 'train/ece': rm.metrics.ExpectedCalibrationError( num_bins=FLAGS.num_bins), 'test/negative_log_likelihood': tf.keras.metrics.Mean(), 'test/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'test/ece': rm.metrics.ExpectedCalibrationError( num_bins=FLAGS.num_bins), 'test/stddev': tf.keras.metrics.Mean(), } if use_validation_set: metrics.update({ 'val/negative_log_likelihood': tf.keras.metrics.Mean(), 'val/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'val/ece': rm.metrics.ExpectedCalibrationError( num_bins=FLAGS.num_bins), 'val/stddev': tf.keras.metrics.Mean(), }) if FLAGS.eval_on_ood: ood_metrics = ood_utils.create_ood_metrics( ood_dataset_names, tpr_list=FLAGS.ood_tpr_threshold) metrics.update(ood_metrics) if FLAGS.corruptions_interval > 0: corrupt_metrics = {} for intensity in range(1, 6): for corruption in corruption_types: dataset_name = '{0}_{1}'.format(corruption, intensity) corrupt_metrics['test/nll_{}'.format(dataset_name)] = ( tf.keras.metrics.Mean()) corrupt_metrics['test/accuracy_{}'.format(dataset_name)] = ( tf.keras.metrics.SparseCategoricalAccuracy()) corrupt_metrics['test/ece_{}'.format(dataset_name)] = ( rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins)) corrupt_metrics['test/stddev_{}'.format(dataset_name)] = ( tf.keras.metrics.Mean()) checkpoint = tf.train.Checkpoint(model=model, optimizer=optimizer) latest_checkpoint = tf.train.latest_checkpoint(FLAGS.output_dir) initial_epoch = 0 logging.info('Output dir : %s', FLAGS.output_dir) if latest_checkpoint: # checkpoint.restore must be within a strategy.scope() so that optimizer # slot variables are mirrored. checkpoint.restore(latest_checkpoint) logging.info('Loaded checkpoint %s', latest_checkpoint) initial_epoch = optimizer.iterations.numpy() // steps_per_epoch if FLAGS.saved_model_dir: logging.info('Saved model dir : %s', FLAGS.saved_model_dir) latest_checkpoint = tf.train.latest_checkpoint(FLAGS.saved_model_dir) checkpoint.restore(latest_checkpoint) logging.info('Loaded checkpoint %s', latest_checkpoint) if FLAGS.eval_only: initial_epoch = FLAGS.train_epochs - 1 # Run just one epoch of eval @tf.function def train_step(iterator, step): """Training StepFn.""" def step_fn(inputs, step): """Per-Replica StepFn.""" images = inputs['features'] labels = inputs['labels'] if tf.equal(step, 0) and FLAGS.gp_cov_discount_factor < 0: # Resetting covaraince estimator at the begining of a new epoch. if FLAGS.use_gp_layer: model.layers[-1].reset_covariance_matrix() if FLAGS.augmix and FLAGS.aug_count >= 1: # Index 0 at augmix preprocessing is the unperturbed image. images = images[:, 1, ...] # This is for the case of combining AugMix and Mixup. if FLAGS.mixup_alpha > 0: labels = tf.split(labels, FLAGS.aug_count + 1, axis=0)[1] images = tf.tile(images, [FLAGS.num_dropout_samples_training, 1, 1, 1]) if FLAGS.mixup_alpha > 0: labels = tf.tile(labels, [FLAGS.num_dropout_samples_training, 1]) else: labels = tf.tile(labels, [FLAGS.num_dropout_samples_training]) with tf.GradientTape() as tape: logits = model(images, training=True) if isinstance(logits, (list, tuple)): # If model returns a tuple of (logits, covmat), extract logits logits, _ = logits if FLAGS.use_bfloat16: logits = tf.cast(logits, tf.float32) if FLAGS.mixup_alpha > 0: negative_log_likelihood = tf.reduce_mean( tf.keras.losses.categorical_crossentropy(labels, logits, from_logits=True)) else: negative_log_likelihood = tf.reduce_mean( tf.keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)) l2_loss = sum(model.losses) loss = negative_log_likelihood + l2_loss # Scale the loss given the TPUStrategy will reduce sum all gradients. scaled_loss = loss / strategy.num_replicas_in_sync grads = tape.gradient(scaled_loss, model.trainable_variables) optimizer.apply_gradients(zip(grads, model.trainable_variables)) probs = tf.nn.softmax(logits) if FLAGS.mixup_alpha > 0: labels = tf.argmax(labels, axis=-1) metrics['train/ece'].add_batch(probs, label=labels) metrics['train/loss'].update_state(loss) metrics['train/negative_log_likelihood'].update_state( negative_log_likelihood) metrics['train/accuracy'].update_state(labels, logits) strategy.run(step_fn, args=(next(iterator), step)) @tf.function def test_step(iterator, dataset_name, num_steps): """Evaluation StepFn.""" def step_fn(inputs): """Per-Replica StepFn.""" images = inputs['features'] labels = inputs['labels'] logits_list = [] stddev_list = [] for _ in range(FLAGS.num_dropout_samples): logits = model(images, training=False) if isinstance(logits, (list, tuple)): # If model returns a tuple of (logits, covmat), extract both logits, covmat = logits if FLAGS.use_bfloat16: logits = tf.cast(logits, tf.float32) logits = ed.layers.utils.mean_field_logits( logits, covmat, mean_field_factor=FLAGS.gp_mean_field_factor) else: covmat = tf.eye(logits.shape[0]) if FLAGS.use_bfloat16: logits = tf.cast(logits, tf.float32) stddev = tf.sqrt(tf.linalg.diag_part(covmat)) stddev_list.append(stddev) logits_list.append(logits) # Logits dimension is (num_samples, batch_size, num_classes). logits_list = tf.stack(logits_list, axis=0) stddev_list = tf.stack(stddev_list, axis=0) stddev = tf.reduce_mean(stddev_list, axis=0) probs_list = tf.nn.softmax(logits_list) probs = tf.reduce_mean(probs_list, axis=0) logits = tf.reduce_mean(logits_list, axis=0) labels_broadcasted = tf.broadcast_to( labels, [FLAGS.num_dropout_samples, tf.shape(labels)[0]]) log_likelihoods = -tf.keras.losses.sparse_categorical_crossentropy( labels_broadcasted, logits_list, from_logits=True) negative_log_likelihood = tf.reduce_mean( -tf.reduce_logsumexp(log_likelihoods, axis=[0]) + tf.math.log(float(FLAGS.num_dropout_samples))) logging.info('Dataset name : %s', dataset_name) if dataset_name == 'clean': metrics['test/negative_log_likelihood'].update_state( negative_log_likelihood) metrics['test/accuracy'].update_state(labels, probs) metrics['test/ece'].add_batch(probs, label=labels) metrics['test/stddev'].update_state(stddev) elif dataset_name == 'val': metrics['val/negative_log_likelihood'].update_state( negative_log_likelihood) metrics['val/accuracy'].update_state(labels, probs) metrics['val/ece'].add_batch(probs, label=labels) metrics['val/stddev'].update_state(stddev) elif dataset_name.startswith('ood/'): ood_labels = 1 - inputs['is_in_distribution'] if FLAGS.dempster_shafer_ood: ood_scores = ood_utils.DempsterShaferUncertainty(logits) else: ood_scores = 1 - tf.reduce_max(probs, axis=-1) # Edgecase for if dataset_name contains underscores for name, metric in metrics.items(): if dataset_name in name: metric.update_state(ood_labels, ood_scores) elif FLAGS.corruptions_interval > 0: corrupt_metrics['test/nll_{}'.format(dataset_name)].update_state( negative_log_likelihood) corrupt_metrics['test/accuracy_{}'.format(dataset_name)].update_state( labels, probs) corrupt_metrics['test/ece_{}'.format(dataset_name)].add_batch( probs, label=labels) corrupt_metrics['test/stddev_{}'.format(dataset_name)].update_state( stddev) for _ in tf.range(tf.cast(num_steps, tf.int32)): strategy.run(step_fn, args=(next(iterator),)) metrics.update({'test/ms_per_example': tf.keras.metrics.Mean()}) step_variable = tf.Variable(0, dtype=tf.int32) train_iterator = iter(train_dataset) start_time = time.time() for epoch in range(initial_epoch, FLAGS.train_epochs): logging.info('Starting to run epoch: %s', epoch) if not FLAGS.eval_only: for step in range(steps_per_epoch): step_variable.assign(step) # Pass `step` as a tf.Variable to train_step to prevent the tf.function # train_step() re-compiling itself at each function call. train_step(train_iterator, step_variable) current_step = epoch * steps_per_epoch + (step + 1) max_steps = steps_per_epoch * FLAGS.train_epochs time_elapsed = time.time() - start_time steps_per_sec = float(current_step) / time_elapsed eta_seconds = (max_steps - current_step) / steps_per_sec message = ('{:.1%} completion: epoch {:d}/{:d}. {:.1f} steps/s. ' 'ETA: {:.0f} min. Time elapsed: {:.0f} min'.format( current_step / max_steps, epoch + 1, FLAGS.train_epochs, steps_per_sec, eta_seconds / 60, time_elapsed / 60)) if step % 20 == 0: logging.info(message) datasets_to_evaluate = {'clean': test_datasets['clean']} if use_validation_set: datasets_to_evaluate['val'] = validation_dataset if (FLAGS.corruptions_interval > 0 and (epoch + 1) % FLAGS.corruptions_interval == 0): datasets_to_evaluate = test_datasets for dataset_name, test_dataset in datasets_to_evaluate.items(): test_iterator = iter(test_dataset) logging.info('Testing on dataset %s', dataset_name) steps_per_eval = steps_per_val if dataset_name == 'val' else steps_per_eval logging.info('Starting to run eval at epoch: %s', epoch) test_start_time = time.time() test_step(test_iterator, dataset_name, steps_per_eval) ms_per_example = (time.time() - test_start_time) * 1e6 / batch_size metrics['test/ms_per_example'].update_state(ms_per_example) logging.info('Done with testing on %s', dataset_name) if FLAGS.eval_on_ood: for ood_dataset_name, ood_dataset in ood_datasets.items(): ood_iterator = iter(ood_dataset) logging.info('Calculating OOD on dataset %s', ood_dataset_name) logging.info('Running OOD eval at epoch: %s', epoch) test_step(ood_iterator, ood_dataset_name, steps_per_ood[ood_dataset_name]) logging.info('Done with OOD eval on %s', dataset_name) corrupt_results = {} if (FLAGS.corruptions_interval > 0 and (epoch + 1) % FLAGS.corruptions_interval == 0): corrupt_results = utils.aggregate_corrupt_metrics(corrupt_metrics, corruption_types) logging.info('Train Loss: %.4f, Accuracy: %.2f%%', metrics['train/loss'].result(), metrics['train/accuracy'].result() * 100) if use_validation_set: logging.info('Val NLL: %.4f, Accuracy: %.2f%%', metrics['val/negative_log_likelihood'].result(), metrics['val/accuracy'].result() * 100) logging.info('Test NLL: %.4f, Accuracy: %.2f%%', metrics['test/negative_log_likelihood'].result(), metrics['test/accuracy'].result() * 100) total_results = {name: metric.result() for name, metric in metrics.items()} total_results.update(corrupt_results) # Metrics from Robustness Metrics (like ECE) will return a dict with a # single key/value, instead of a scalar. total_results = { k: (list(v.values())[0] if isinstance(v, dict) else v) for k, v in total_results.items() } with summary_writer.as_default(): for name, result in total_results.items(): tf.summary.scalar(name, result, step=epoch + 1) for metric in metrics.values(): metric.reset_states() if FLAGS.corruptions_interval > 0: for metric in corrupt_metrics.values(): metric.reset_states() if (FLAGS.checkpoint_interval > 0 and (epoch + 1) % FLAGS.checkpoint_interval == 0): checkpoint_name = checkpoint.save( os.path.join(FLAGS.output_dir, 'checkpoint')) logging.info('Saved checkpoint to %s', checkpoint_name) final_checkpoint_name = checkpoint.save( os.path.join(FLAGS.output_dir, 'checkpoint')) logging.info('Saved last checkpoint to %s', final_checkpoint_name) final_save_name = os.path.join(FLAGS.output_dir, 'model') model.save(final_save_name) logging.info('Saved model to %s', final_save_name) with summary_writer.as_default(): hp.hparams({ 'base_learning_rate': FLAGS.base_learning_rate, 'one_minus_momentum': FLAGS.one_minus_momentum, 'l2': FLAGS.l2, 'gp_mean_field_factor': FLAGS.gp_mean_field_factor, })
def main(argv): del argv # unused arg tf.io.gfile.makedirs(FLAGS.output_dir) logging.info('Saving checkpoints at %s', FLAGS.output_dir) tf.random.set_seed(FLAGS.seed) data_dir = FLAGS.data_dir if FLAGS.use_gpu: logging.info('Use GPU') strategy = tf.distribute.MirroredStrategy() else: logging.info('Use TPU at %s', FLAGS.tpu if FLAGS.tpu is not None else 'local') resolver = tf.distribute.cluster_resolver.TPUClusterResolver( tpu=FLAGS.tpu) tf.config.experimental_connect_to_cluster(resolver) tf.tpu.experimental.initialize_tpu_system(resolver) strategy = tf.distribute.TPUStrategy(resolver) ds_info = tfds.builder(FLAGS.dataset).info batch_size = (FLAGS.per_core_batch_size * FLAGS.num_cores // FLAGS.num_dropout_samples_training) test_batch_size = FLAGS.per_core_batch_size * FLAGS.num_cores steps_per_epoch = ds_info.splits['train'].num_examples // batch_size steps_per_eval = ds_info.splits['test'].num_examples // test_batch_size num_classes = ds_info.features['label'].num_classes train_builder = ub.datasets.get(FLAGS.dataset, data_dir=data_dir, download_data=FLAGS.download_data, split=tfds.Split.TRAIN, validation_percent=1. - FLAGS.train_proportion) train_dataset = train_builder.load(batch_size=batch_size) validation_dataset = None steps_per_validation = 0 if FLAGS.train_proportion < 1.0: validation_builder = ub.datasets.get( FLAGS.dataset, data_dir=data_dir, split=tfds.Split.VALIDATION, validation_percent=1. - FLAGS.train_proportion, drop_remainder=FLAGS.drop_remainder_for_eval) validation_dataset = validation_builder.load( batch_size=test_batch_size) validation_dataset = strategy.experimental_distribute_dataset( validation_dataset) steps_per_validation = validation_builder.num_examples // test_batch_size clean_test_builder = ub.datasets.get( FLAGS.dataset, data_dir=data_dir, split=tfds.Split.TEST, drop_remainder=FLAGS.drop_remainder_for_eval) clean_test_dataset = clean_test_builder.load(batch_size=test_batch_size) train_dataset = strategy.experimental_distribute_dataset(train_dataset) test_datasets = { 'clean': strategy.experimental_distribute_dataset(clean_test_dataset), } steps_per_epoch = train_builder.num_examples // batch_size steps_per_eval = clean_test_builder.num_examples // batch_size num_classes = 100 if FLAGS.dataset == 'cifar100' else 10 if FLAGS.eval_on_ood: ood_dataset_names = FLAGS.ood_dataset ood_ds, steps_per_ood = ood_utils.load_ood_datasets( ood_dataset_names, clean_test_builder, 1. - FLAGS.train_proportion, batch_size, drop_remainder=FLAGS.drop_remainder_for_eval) ood_datasets = { name: strategy.experimental_distribute_dataset(ds) for name, ds in ood_ds.items() } if FLAGS.corruptions_interval > 0: if FLAGS.dataset == 'cifar100': data_dir = FLAGS.cifar100_c_path corruption_types, _ = utils.load_corrupted_test_info(FLAGS.dataset) for corruption_type in corruption_types: for severity in range(1, 6): dataset = ub.datasets.get( f'{FLAGS.dataset}_corrupted', corruption_type=corruption_type, data_dir=data_dir, severity=severity, split=tfds.Split.TEST, drop_remainder=FLAGS.drop_remainder_for_eval).load( batch_size=test_batch_size) test_datasets[f'{corruption_type}_{severity}'] = ( strategy.experimental_distribute_dataset(dataset)) summary_writer = tf.summary.create_file_writer( os.path.join(FLAGS.output_dir, 'summaries')) with strategy.scope(): logging.info('Building ResNet model') model = ub.models.wide_resnet_dropout( input_shape=(32, 32, 3), depth=28, width_multiplier=10, num_classes=num_classes, l2=FLAGS.l2, dropout_rate=FLAGS.dropout_rate, residual_dropout=FLAGS.residual_dropout, filterwise_dropout=FLAGS.filterwise_dropout) logging.info('Model input shape: %s', model.input_shape) logging.info('Model output shape: %s', model.output_shape) logging.info('Model number of weights: %s', model.count_params()) # Linearly scale learning rate and the decay epochs by vanilla settings. base_lr = FLAGS.base_learning_rate * batch_size / 128 lr_decay_epochs = [(int(start_epoch_str) * FLAGS.train_epochs) // 200 for start_epoch_str in FLAGS.lr_decay_epochs] lr_schedule = ub.schedules.WarmUpPiecewiseConstantSchedule( steps_per_epoch, base_lr, decay_ratio=FLAGS.lr_decay_ratio, decay_epochs=lr_decay_epochs, warmup_epochs=FLAGS.lr_warmup_epochs) optimizer = tf.keras.optimizers.SGD(lr_schedule, momentum=1.0 - FLAGS.one_minus_momentum, nesterov=True) metrics = { 'train/negative_log_likelihood': tf.keras.metrics.Mean(), 'train/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'train/loss': tf.keras.metrics.Mean(), 'train/ece': rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins), 'test/negative_log_likelihood': tf.keras.metrics.Mean(), 'test/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'test/ece': rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins), } if validation_dataset: metrics.update({ 'validation/negative_log_likelihood': tf.keras.metrics.Mean(), 'validation/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'validation/ece': rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins), }) if FLAGS.eval_on_ood: ood_metrics = ood_utils.create_ood_metrics(ood_dataset_names) metrics.update(ood_metrics) if FLAGS.corruptions_interval > 0: corrupt_metrics = {} for intensity in range(1, 6): for corruption in corruption_types: dataset_name = '{0}_{1}'.format(corruption, intensity) corrupt_metrics['test/nll_{}'.format(dataset_name)] = ( tf.keras.metrics.Mean()) corrupt_metrics['test/accuracy_{}'.format( dataset_name)] = ( tf.keras.metrics.SparseCategoricalAccuracy()) corrupt_metrics['test/ece_{}'.format(dataset_name)] = ( rm.metrics.ExpectedCalibrationError( num_bins=FLAGS.num_bins)) checkpoint = tf.train.Checkpoint(model=model, optimizer=optimizer) latest_checkpoint = tf.train.latest_checkpoint(FLAGS.output_dir) initial_epoch = 0 if latest_checkpoint: # checkpoint.restore must be within a strategy.scope() so that optimizer # slot variables are mirrored. checkpoint.restore(latest_checkpoint) logging.info('Loaded checkpoint %s', latest_checkpoint) initial_epoch = optimizer.iterations.numpy() // steps_per_epoch @tf.function def train_step(iterator): """Training StepFn.""" def step_fn(inputs): """Per-Replica StepFn.""" images = inputs['features'] labels = inputs['labels'] images = tf.tile(images, [FLAGS.num_dropout_samples_training, 1, 1, 1]) labels = tf.tile(labels, [FLAGS.num_dropout_samples_training]) with tf.GradientTape() as tape: logits = model(images, training=True) negative_log_likelihood = tf.reduce_mean( tf.keras.losses.sparse_categorical_crossentropy( labels, logits, from_logits=True)) l2_loss = sum(model.losses) loss = negative_log_likelihood + l2_loss # Scale the loss given the TPUStrategy will reduce sum all gradients. scaled_loss = loss / strategy.num_replicas_in_sync grads = tape.gradient(scaled_loss, model.trainable_variables) optimizer.apply_gradients(zip(grads, model.trainable_variables)) probs = tf.nn.softmax(logits) metrics['train/ece'].add_batch(probs, label=labels) metrics['train/loss'].update_state(loss) metrics['train/negative_log_likelihood'].update_state( negative_log_likelihood) metrics['train/accuracy'].update_state(labels, logits) for _ in tf.range(tf.cast(steps_per_epoch, tf.int32)): strategy.run(step_fn, args=(next(iterator), )) @tf.function def test_step(iterator, dataset_split, dataset_name, num_steps): """Evaluation StepFn.""" def step_fn(inputs): """Per-Replica StepFn.""" images = inputs['features'] labels = inputs['labels'] logits_list = [] for _ in range(FLAGS.num_dropout_samples): logits = model(images, training=False) logits_list.append(logits) # Logits dimension is (num_samples, batch_size, num_classes). logits_list = tf.stack(logits_list, axis=0) probs_list = tf.nn.softmax(logits_list) probs = tf.reduce_mean(probs_list, axis=0) labels_broadcasted = tf.broadcast_to( labels, [FLAGS.num_dropout_samples, tf.shape(labels)[0]]) log_likelihoods = -tf.keras.losses.sparse_categorical_crossentropy( labels_broadcasted, logits_list, from_logits=True) negative_log_likelihood = tf.reduce_mean( -tf.reduce_logsumexp(log_likelihoods, axis=[0]) + tf.math.log(float(FLAGS.num_dropout_samples))) if dataset_name == 'clean': metrics[ f'{dataset_split}/negative_log_likelihood'].update_state( negative_log_likelihood) metrics[f'{dataset_split}/accuracy'].update_state( labels, probs) metrics[f'{dataset_split}/ece'].add_batch(probs, label=labels) elif dataset_name.startswith('ood/'): ood_labels = 1 - inputs['is_in_distribution'] if FLAGS.dempster_shafer_ood: ood_scores = ood_utils.DempsterShaferUncertainty(logits) else: ood_scores = 1 - tf.reduce_max(probs, axis=-1) for name, metric in metrics.items(): if dataset_name in name: metric.update_state(ood_labels, ood_scores) else: corrupt_metrics['test/nll_{}'.format( dataset_name)].update_state(negative_log_likelihood) corrupt_metrics['test/accuracy_{}'.format( dataset_name)].update_state(labels, probs) corrupt_metrics['test/ece_{}'.format(dataset_name)].add_batch( probs, label=labels) for _ in tf.range(tf.cast(num_steps, tf.int32)): strategy.run(step_fn, args=(next(iterator), )) metrics.update({'test/ms_per_example': tf.keras.metrics.Mean()}) train_iterator = iter(train_dataset) start_time = time.time() for epoch in range(initial_epoch, FLAGS.train_epochs): logging.info('Starting to run epoch: %s', epoch) train_step(train_iterator) current_step = (epoch + 1) * steps_per_epoch max_steps = steps_per_epoch * FLAGS.train_epochs time_elapsed = time.time() - start_time steps_per_sec = float(current_step) / time_elapsed eta_seconds = (max_steps - current_step) / steps_per_sec message = ('{:.1%} completion: epoch {:d}/{:d}. {:.1f} steps/s. ' 'ETA: {:.0f} min. Time elapsed: {:.0f} min'.format( current_step / max_steps, epoch + 1, FLAGS.train_epochs, steps_per_sec, eta_seconds / 60, time_elapsed / 60)) logging.info(message) if validation_dataset: validation_iterator = iter(validation_dataset) test_step(validation_iterator, 'validation', 'clean', steps_per_validation) datasets_to_evaluate = {'clean': test_datasets['clean']} if (FLAGS.corruptions_interval > 0 and (epoch + 1) % FLAGS.corruptions_interval == 0): datasets_to_evaluate = test_datasets for dataset_name, test_dataset in datasets_to_evaluate.items(): test_iterator = iter(test_dataset) logging.info('Testing on dataset %s', dataset_name) logging.info('Starting to run eval at epoch: %s', epoch) test_start_time = time.time() test_step(test_iterator, 'test', dataset_name, steps_per_eval) ms_per_example = (time.time() - test_start_time) * 1e6 / batch_size metrics['test/ms_per_example'].update_state(ms_per_example) logging.info('Done with testing on %s', dataset_name) if FLAGS.eval_on_ood: for ood_dataset_name, ood_dataset in ood_datasets.items(): ood_iterator = iter(ood_dataset) logging.info('Calculating OOD on dataset %s', ood_dataset_name) logging.info('Running OOD eval at epoch: %s', epoch) test_step(ood_iterator, 'test', ood_dataset_name, steps_per_ood[ood_dataset_name]) logging.info('Done with OOD eval on %s', dataset_name) corrupt_results = {} if (FLAGS.corruptions_interval > 0 and (epoch + 1) % FLAGS.corruptions_interval == 0): corrupt_results = utils.aggregate_corrupt_metrics( corrupt_metrics, corruption_types) logging.info('Train Loss: %.4f, Accuracy: %.2f%%', metrics['train/loss'].result(), metrics['train/accuracy'].result() * 100) logging.info('Test NLL: %.4f, Accuracy: %.2f%%', metrics['test/negative_log_likelihood'].result(), metrics['test/accuracy'].result() * 100) total_results = { name: metric.result() for name, metric in metrics.items() } total_results.update(corrupt_results) # Metrics from Robustness Metrics (like ECE) will return a dict with a # single key/value, instead of a scalar. total_results = { k: (list(v.values())[0] if isinstance(v, dict) else v) for k, v in total_results.items() } with summary_writer.as_default(): for name, result in total_results.items(): tf.summary.scalar(name, result, step=epoch + 1) for metric in metrics.values(): metric.reset_states() if FLAGS.corruptions_interval > 0: for metric in corrupt_metrics.values(): metric.reset_states() if (FLAGS.checkpoint_interval > 0 and (epoch + 1) % FLAGS.checkpoint_interval == 0): checkpoint_name = checkpoint.save( os.path.join(FLAGS.output_dir, 'checkpoint')) logging.info('Saved checkpoint to %s', checkpoint_name) final_checkpoint_name = checkpoint.save( os.path.join(FLAGS.output_dir, 'checkpoint')) logging.info('Saved last checkpoint to %s', final_checkpoint_name) with summary_writer.as_default(): hp.hparams({ 'base_learning_rate': FLAGS.base_learning_rate, 'one_minus_momentum': FLAGS.one_minus_momentum, 'l2': FLAGS.l2, 'dropout_rate': FLAGS.dropout_rate, 'num_dropout_samples': FLAGS.num_dropout_samples, 'num_dropout_samples_training': FLAGS.num_dropout_samples_training, })
def main(argv): del argv # unused arg if not FLAGS.use_gpu: raise ValueError('Only GPU is currently supported.') if FLAGS.num_cores > 1: raise ValueError('Only a single accelerator is currently supported.') tf.random.set_seed(FLAGS.seed) tf.io.gfile.makedirs(FLAGS.output_dir) logging.info('output_dir=%s', FLAGS.output_dir) ds_info = tfds.builder(FLAGS.dataset).info batch_size = FLAGS.per_core_batch_size * FLAGS.num_cores steps_per_eval = ds_info.splits['test'].num_examples // batch_size num_classes = ds_info.features['label'].num_classes data_dir = FLAGS.data_dir dataset_builder = ub.datasets.get( FLAGS.dataset, download_data=FLAGS.download_data, data_dir=data_dir, split=tfds.Split.TEST, drop_remainder=FLAGS.drop_remainder_for_eval) dataset = dataset_builder.load(batch_size=batch_size) test_datasets = {'clean': dataset} if FLAGS.eval_on_ood: ood_dataset_names = FLAGS.ood_dataset ood_datasets, steps_per_ood = ood_utils.load_ood_datasets( ood_dataset_names, dataset_builder, 1. - FLAGS.train_proportion, batch_size, drop_remainder=FLAGS.drop_remainder_for_eval) test_datasets.update(ood_datasets) if FLAGS.dataset == 'cifar100': data_dir = FLAGS.cifar100_c_path corruption_types, _ = utils.load_corrupted_test_info(FLAGS.dataset) for corruption_type in corruption_types: for severity in range(1, 6): dataset = ub.datasets.get( f'{FLAGS.dataset}_corrupted', corruption_type=corruption_type, download_data=FLAGS.download_data, data_dir=data_dir, severity=severity, split=tfds.Split.TEST, drop_remainder=FLAGS.drop_remainder_for_eval).load( batch_size=batch_size) test_datasets[f'{corruption_type}_{severity}'] = dataset model = ub.models.wide_resnet(input_shape=ds_info.features['image'].shape, depth=28, width_multiplier=10, num_classes=num_classes, l2=0., version=2) logging.info('Model input shape: %s', model.input_shape) logging.info('Model output shape: %s', model.output_shape) logging.info('Model number of weights: %s', model.count_params()) # Search for checkpoints from their index file; then remove the index suffix. ensemble_filenames = parse_checkpoint_dir(FLAGS.checkpoint_dir) ensemble_size = len(ensemble_filenames) logging.info('Ensemble size: %s', ensemble_size) logging.info('Ensemble number of weights: %s', ensemble_size * model.count_params()) logging.info('Ensemble filenames: %s', str(ensemble_filenames)) checkpoint = tf.train.Checkpoint(model=model) # Write model predictions to files. num_datasets = len(test_datasets) for m, ensemble_filename in enumerate(ensemble_filenames): checkpoint.restore(ensemble_filename) for n, (name, test_dataset) in enumerate(test_datasets.items()): filename = '{dataset}_{member}.npy'.format( dataset=name.replace('/', '_'), member=m) # ood dataset name has '/' filename = os.path.join(FLAGS.output_dir, filename) if not tf.io.gfile.exists(filename): logits = [] test_iterator = iter(test_dataset) steps = steps_per_eval if 'ood/' not in name else steps_per_ood[ name] for _ in range(steps): features = next(test_iterator)['features'] # pytype: disable=unsupported-operands logits.append(model(features, training=False)) logits = tf.concat(logits, axis=0) with tf.io.gfile.GFile(filename, 'w') as f: np.save(f, logits.numpy()) percent = (m * num_datasets + (n + 1)) / (ensemble_size * num_datasets) message = ( '{:.1%} completion for prediction: ensemble member {:d}/{:d}. ' 'Dataset {:d}/{:d}'.format(percent, m + 1, ensemble_size, n + 1, num_datasets)) logging.info(message) metrics = { 'test/negative_log_likelihood': tf.keras.metrics.Mean(), 'test/gibbs_cross_entropy': tf.keras.metrics.Mean(), 'test/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'test/ece': rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins), 'test/diversity': rm.metrics.AveragePairwiseDiversity(), } if FLAGS.eval_on_ood: ood_metrics = ood_utils.create_ood_metrics(ood_dataset_names) metrics.update(ood_metrics) corrupt_metrics = {} for name in test_datasets: corrupt_metrics['test/nll_{}'.format(name)] = tf.keras.metrics.Mean() corrupt_metrics['test/accuracy_{}'.format(name)] = ( tf.keras.metrics.SparseCategoricalAccuracy()) corrupt_metrics['test/ece_{}'.format(name)] = ( rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins)) for i in range(ensemble_size): metrics['test/nll_member_{}'.format(i)] = tf.keras.metrics.Mean() metrics['test/accuracy_member_{}'.format(i)] = ( tf.keras.metrics.SparseCategoricalAccuracy()) # Evaluate model predictions. for n, (name, test_dataset) in enumerate(test_datasets.items()): logits_dataset = [] for m in range(ensemble_size): filename = '{dataset}_{member}.npy'.format( dataset=name.replace('/', '_'), member=m) # ood dataset name has '/' filename = os.path.join(FLAGS.output_dir, filename) with tf.io.gfile.GFile(filename, 'rb') as f: logits_dataset.append(np.load(f)) logits_dataset = tf.convert_to_tensor(logits_dataset) test_iterator = iter(test_dataset) steps = steps_per_eval if 'ood/' not in name else steps_per_ood[name] for step in range(steps): inputs = next(test_iterator) labels = inputs['labels'] # pytype: disable=unsupported-operands logits = logits_dataset[:, (step * batch_size):((step + 1) * batch_size)] labels = tf.cast(labels, tf.int32) negative_log_likelihood_metric = rm.metrics.EnsembleCrossEntropy() negative_log_likelihood_metric.add_batch(logits, labels=labels) negative_log_likelihood = list( negative_log_likelihood_metric.result().values())[0] per_probs = tf.nn.softmax(logits) probs = tf.reduce_mean(per_probs, axis=0) logits_mean = tf.reduce_mean(logits, axis=0) if name == 'clean': gibbs_ce_metric = rm.metrics.GibbsCrossEntropy() gibbs_ce_metric.add_batch(logits, labels=labels) gibbs_ce = list(gibbs_ce_metric.result().values())[0] metrics['test/negative_log_likelihood'].update_state( negative_log_likelihood) metrics['test/gibbs_cross_entropy'].update_state(gibbs_ce) metrics['test/accuracy'].update_state(labels, probs) metrics['test/ece'].add_batch(probs, label=labels) for i in range(ensemble_size): member_probs = per_probs[i] member_loss = tf.keras.losses.sparse_categorical_crossentropy( labels, member_probs) metrics['test/nll_member_{}'.format(i)].update_state( member_loss) metrics['test/accuracy_member_{}'.format(i)].update_state( labels, member_probs) metrics['test/diversity'].add_batch(per_probs) elif name.startswith('ood/'): ood_labels = 1 - inputs['is_in_distribution'] # pytype: disable=unsupported-operands if FLAGS.dempster_shafer_ood: ood_scores = ood_utils.DempsterShaferUncertainty( logits_mean) else: ood_scores = 1 - tf.reduce_max(probs, axis=-1) for metric_name, metric in metrics.items(): if name in metric_name: metric.update_state(ood_labels, ood_scores) else: corrupt_metrics['test/nll_{}'.format(name)].update_state( negative_log_likelihood) corrupt_metrics['test/accuracy_{}'.format(name)].update_state( labels, probs) corrupt_metrics['test/ece_{}'.format(name)].add_batch( probs, label=labels) message = ( '{:.1%} completion for evaluation: dataset {:d}/{:d}'.format( (n + 1) / num_datasets, n + 1, num_datasets)) logging.info(message) corrupt_results = utils.aggregate_corrupt_metrics(corrupt_metrics, corruption_types) total_results = {name: metric.result() for name, metric in metrics.items()} total_results.update(corrupt_results) # Results from Robustness Metrics themselves return a dict, so flatten them. total_results = utils.flatten_dictionary(total_results) logging.info('Metrics: %s', total_results)
def main(argv): del argv # unused arg if not FLAGS.use_gpu: raise ValueError('Only GPU is currently supported.') if FLAGS.num_cores > 1: raise ValueError('Only a single accelerator is currently supported.') tf.random.set_seed(FLAGS.seed) tf.io.gfile.makedirs(FLAGS.output_dir) ds_info = tfds.builder(FLAGS.dataset).info batch_size = FLAGS.total_batch_size steps_per_eval = ds_info.splits['test'].num_examples // batch_size num_classes = ds_info.features['label'].num_classes data_dir = FLAGS.data_dir dataset_builder = ub.datasets.get( FLAGS.dataset, data_dir=data_dir, download_data=FLAGS.download_data, split=tfds.Split.TEST, drop_remainder=FLAGS.drop_remainder_for_eval) dataset = dataset_builder.load(batch_size=batch_size) test_datasets = {'clean': dataset} if FLAGS.eval_on_ood: ood_dataset_names = FLAGS.ood_dataset ood_datasets, steps_per_ood = ood_utils.load_ood_datasets( ood_dataset_names, dataset_builder, 1. - FLAGS.train_proportion, batch_size, drop_remainder=FLAGS.drop_remainder_for_eval) test_datasets.update(ood_datasets) extra_kwargs = {} if FLAGS.dataset == 'cifar100': data_dir = FLAGS.cifar100_c_path corruption_types, _ = utils.load_corrupted_test_info(FLAGS.dataset) for corruption_type in corruption_types: for severity in range(1, 6): dataset = ub.datasets.get( f'{FLAGS.dataset}_corrupted', corruption_type=corruption_type, data_dir=data_dir, severity=severity, split=tfds.Split.TEST, drop_remainder=FLAGS.drop_remainder_for_eval, **extra_kwargs).load(batch_size=batch_size) test_datasets[f'{corruption_type}_{severity}'] = dataset model = ub.models.wide_resnet_sngp( input_shape=ds_info.features['image'].shape, batch_size=FLAGS.total_batch_size // FLAGS.num_cores, depth=28, width_multiplier=10, num_classes=num_classes, l2=0., use_mc_dropout=FLAGS.use_mc_dropout, use_filterwise_dropout=FLAGS.use_filterwise_dropout, dropout_rate=FLAGS.dropout_rate, use_gp_layer=FLAGS.use_gp_layer, gp_input_dim=FLAGS.gp_input_dim, gp_hidden_dim=FLAGS.gp_hidden_dim, gp_scale=FLAGS.gp_scale, gp_bias=FLAGS.gp_bias, gp_input_normalization=FLAGS.gp_input_normalization, gp_random_feature_type=FLAGS.gp_random_feature_type, gp_cov_discount_factor=FLAGS.gp_cov_discount_factor, gp_cov_ridge_penalty=FLAGS.gp_cov_ridge_penalty, use_spec_norm=FLAGS.use_spec_norm, spec_norm_iteration=FLAGS.spec_norm_iteration, spec_norm_bound=FLAGS.spec_norm_bound) logging.info('Model input shape: %s', model.input_shape) logging.info('Model output shape: %s', model.output_shape) logging.info('Model number of weights: %s', model.count_params()) # Search for checkpoints from their index file; then remove the index suffix. ensemble_filenames = tf.io.gfile.glob( os.path.join(FLAGS.checkpoint_dir, '**/*.index')) # Only apply ensemble on the models with the same model architecture ensemble_filenames0 = [ filename for filename in ensemble_filenames if f'use_gp_layer:{FLAGS.use_gp_layer}' in filename and f'use_spec_norm:{FLAGS.use_spec_norm}' in filename ] np.random.seed(FLAGS.seed) ensemble_filenames = np.random.choice(ensemble_filenames0, FLAGS.ensemble_size, replace=True) ensemble_filenames = [filename[:-6] for filename in ensemble_filenames] ensemble_size = len(ensemble_filenames) logging.info('Ensemble size: %s', ensemble_size) logging.info('Ensemble filenames: %s', ensemble_filenames) logging.info('Ensemble number of weights: %s', ensemble_size * model.count_params()) logging.info('Ensemble filenames: %s', str(ensemble_filenames)) checkpoint = tf.train.Checkpoint(model=model) # Write model predictions to files. num_datasets = len(test_datasets) for m, ensemble_filename in enumerate(ensemble_filenames): checkpoint.restore(ensemble_filename) for n, (name, test_dataset) in enumerate(test_datasets.items()): filename = '{dataset}_{member}.npy'.format( dataset=name.replace('/', '_'), member=m) # ood dataset name has '/' filename = os.path.join(FLAGS.output_dir, filename) if not tf.io.gfile.exists(filename): logits = [] test_iterator = iter(test_dataset) steps = steps_per_eval if 'ood/' not in name else steps_per_ood[ name] for _ in range(steps): features = next(test_iterator)['features'] # pytype: disable=unsupported-operands logits_member = model(features, training=False) if isinstance(logits_member, (list, tuple)): # If model returns a tuple of (logits, covmat), extract both logits_member, covmat_member = logits_member logits_member = ed.layers.utils.mean_field_logits( logits_member, covmat_member, FLAGS.gp_mean_field_factor_ensemble) logits.append(logits_member) logits = tf.concat(logits, axis=0) with tf.io.gfile.GFile(filename, 'w') as f: np.save(f, logits.numpy()) percent = (m * num_datasets + (n + 1)) / (ensemble_size * num_datasets) message = ( '{:.1%} completion for prediction: ensemble member {:d}/{:d}. ' 'Dataset {:d}/{:d}'.format(percent, m + 1, ensemble_size, n + 1, num_datasets)) logging.info(message) metrics = { 'test/negative_log_likelihood': tf.keras.metrics.Mean(), 'test/gibbs_cross_entropy': tf.keras.metrics.Mean(), 'test/accuracy': tf.keras.metrics.SparseCategoricalAccuracy(), 'test/ece': rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins), } if FLAGS.eval_on_ood: ood_metrics = ood_utils.create_ood_metrics(ood_dataset_names) metrics.update(ood_metrics) corrupt_metrics = {} for name in test_datasets: corrupt_metrics['test/nll_{}'.format(name)] = tf.keras.metrics.Mean() corrupt_metrics['test/accuracy_{}'.format(name)] = ( tf.keras.metrics.SparseCategoricalAccuracy()) corrupt_metrics['test/ece_{}'.format(name)] = ( rm.metrics.ExpectedCalibrationError(num_bins=FLAGS.num_bins)) # Evaluate model predictions. for n, (name, test_dataset) in enumerate(test_datasets.items()): logits_dataset = [] for m in range(ensemble_size): filename = '{dataset}_{member}.npy'.format( dataset=name.replace('/', '_'), member=m) # ood dataset name has '/' filename = os.path.join(FLAGS.output_dir, filename) with tf.io.gfile.GFile(filename, 'rb') as f: logits_dataset.append(np.load(f)) logits_dataset = tf.convert_to_tensor(logits_dataset) test_iterator = iter(test_dataset) steps = steps_per_eval if 'ood/' not in name else steps_per_ood[name] for step in range(steps): inputs = next(test_iterator) labels = inputs['labels'] # pytype: disable=unsupported-operands logits = logits_dataset[:, (step * batch_size):((step + 1) * batch_size)] labels = tf.cast(labels, tf.int32) negative_log_likelihood_metric = rm.metrics.EnsembleCrossEntropy() negative_log_likelihood_metric.add_batch(logits, labels=labels) negative_log_likelihood = list( negative_log_likelihood_metric.result().values())[0] per_probs = tf.nn.softmax(logits) probs = tf.reduce_mean(per_probs, axis=0) logits_mean = tf.reduce_mean(logits, axis=0) if name == 'clean': gibbs_ce_metric = rm.metrics.GibbsCrossEntropy() gibbs_ce_metric.add_batch(logits, labels=labels) gibbs_ce = list(gibbs_ce_metric.result().values())[0] metrics['test/negative_log_likelihood'].update_state( negative_log_likelihood) metrics['test/gibbs_cross_entropy'].update_state(gibbs_ce) metrics['test/accuracy'].update_state(labels, probs) metrics['test/ece'].add_batch(probs, label=labels) elif name.startswith('ood/'): ood_labels = 1 - inputs['is_in_distribution'] # pytype: disable=unsupported-operands if FLAGS.dempster_shafer_ood: ood_scores = ood_utils.DempsterShaferUncertainty( logits_mean) else: ood_scores = 1 - tf.reduce_max(probs, axis=-1) for metric_name, metric in metrics.items(): if name in metric_name: metric.update_state(ood_labels, ood_scores) else: corrupt_metrics['test/nll_{}'.format(name)].update_state( negative_log_likelihood) corrupt_metrics['test/accuracy_{}'.format(name)].update_state( labels, probs) corrupt_metrics['test/ece_{}'.format(name)].add_batch( probs, label=labels) message = ( '{:.1%} completion for evaluation: dataset {:d}/{:d}'.format( (n + 1) / num_datasets, n + 1, num_datasets)) logging.info(message) corrupt_results = utils.aggregate_corrupt_metrics(corrupt_metrics, corruption_types) total_results = {name: metric.result() for name, metric in metrics.items()} total_results.update(corrupt_results) # Metrics from Robustness Metrics (like ECE) will return a dict with a # single key/value, instead of a scalar. total_results = { k: (list(v.values())[0] if isinstance(v, dict) else v) for k, v in total_results.items() } logging.info('Metrics: %s', total_results)