.. _sec_imgadvanced: Image Prediction - Search Space and Hyperparameter Optimization (HPO) ===================================================================== While the :ref:`sec_imgquick` introduced basic usage of AutoGluon ``fit``, ``evaluate``, ``predict`` with default configurations, this tutorial dives into the various options that you can specify for more advanced control over the fitting process. These options include: - Defining the search space of various hyperparameter values for the training of neural networks - Specifying how to search through your choosen hyperparameter space - Specifying how to schedule jobs to train a network under a particular hyperparameter configuration. The advanced functionalities of AutoGluon enable you to use your external knowledge about your particular prediction problem and computing resources to guide the training process. If properly used, you may be able to achieve superior performance within less training time. **Tip**: If you are new to AutoGluon, review :ref:`sec_imgquick` to learn the basics of the AutoGluon API. Since our task is to classify images, we will use AutoGluon to produce an `ImagePredictor <../../api/autogluon.predictor.html#autogluon.vision.ImagePredictor>`__: .. code:: python import autogluon.core as ag from autogluon.vision import ImagePredictor Create AutoGluon Dataset ------------------------ Let's first create the dataset using the same subset of the ``Shopee-IET`` dataset as the :ref:`sec_imgquick` tutorial. Recall that there's no validation split in original data, a 90/10 train/validation split is automatically performed when ``fit`` with ``train_data``. .. code:: python train_data, _, test_data = ImagePredictor.Dataset.from_folders('https://autogluon.s3.amazonaws.com/datasets/shopee-iet.zip') .. parsed-literal:: :class: output data/ ├── test/ └── train/ Specify which Networks to Try ----------------------------- We start with specifying the pretrained neural network candidates. Given such a list, AutoGluon tries to train different networks from this list to identify the best-performing candidate. This is an example of a :class:`autogluon.core.space.Categorical` search space, in which there are a limited number of values to choose from. .. code:: python model = ag.Categorical('resnet18_v1b', 'mobilenetv3_small') # you may choose more than 70+ available model in the model zoo provided by GluonCV: model_list = ImagePredictor.list_models() Specify the training hyper-parameters ------------------------------------- Similarly, we can manually specify many crucial hyper-parameters, with specific value or search space(\ ``autogluon.core.space``). .. code:: python batch_size = 8 lr = ag.Categorical(1e-2, 1e-3) Search Algorithms ----------------- In AutoGluon, ``autogluon.core.searcher`` supports different search search strategies for both hyperparameter optimization and architecture search. Beyond simply specifying the space of hyperparameter configurations to search over, you can also tell AutoGluon what strategy it should employ to actually search through this space. This process of finding good hyperparameters from a given search space is commonly referred to as *hyperparameter optimization* (HPO) or *hyperparameter tuning*. ``autogluon.core.scheduler`` orchestrates how individual training jobs are scheduled. We currently support FIFO (standard) and Hyperband scheduling, along with search by random sampling or Bayesian optimization. These basic techniques are rendered surprisingly powerful by AutoGluon's support of asynchronous parallel execution. Bayesian Optimization ~~~~~~~~~~~~~~~~~~~~~ Here is an example of using Bayesian Optimization using :class:`autogluon.core.searcher.GPFIFOSearcher`. Bayesian Optimization fits a probabilistic *surrogate model* to estimate the function that relates each hyperparameter configuration to the resulting performance of a model trained under this hyperparameter configuration. Our implementation makes use of a Gaussian process surrogate model along with expected improvement as acquisition function. It has been developed specifically to support asynchronous parallel evaluations. .. code:: python hyperparameters={'model': model, 'batch_size': batch_size, 'lr': lr, 'epochs': 2} predictor = ImagePredictor() predictor.fit(train_data, search_strategy='bayesopt', time_limit=60*10, hyperparameters=hyperparameters, hyperparameter_tune_kwargs={'num_trials': 2}) print('Top-1 val acc: %.3f' % predictor.fit_summary()['valid_acc']) .. parsed-literal:: :class: output WARNING:gluoncv.auto.tasks.image_classification:The number of requested GPUs is greater than the number of available GPUs.Reduce the number to 1 INFO:gluoncv.auto.tasks.image_classification:Randomly split train_data into train[729]/validation[71] splits. INFO:gluoncv.auto.tasks.image_classification:Starting HPO experiments INFO:autogluon.core.scheduler.fifo:Starting Hyperparameter Tuning ... (time_out=600s) .. parsed-literal:: :class: output scheduler: FIFOScheduler( DistributedResourceManager{ (Remote: Remote REMOTE_ID: 0, , Resource: NodeResourceManager(8 CPUs, 1 GPUs)) }) .. parsed-literal:: :class: output 0%| | 0/2 [00:00 != ): { INFO:ImageClassificationEstimator:root.train.rec_train_idx ~/.mxnet/datasets/imagenet/rec/train.idx != auto INFO:ImageClassificationEstimator:root.train.data_dir ~/.mxnet/datasets/imagenet != auto INFO:ImageClassificationEstimator:root.train.lr 0.1 != 0.01 INFO:ImageClassificationEstimator:root.train.rec_train ~/.mxnet/datasets/imagenet/rec/train.rec != auto INFO:ImageClassificationEstimator:root.train.epochs 10 != 2 INFO:ImageClassificationEstimator:root.train.rec_val_idx ~/.mxnet/datasets/imagenet/rec/val.idx != auto INFO:ImageClassificationEstimator:root.train.rec_val ~/.mxnet/datasets/imagenet/rec/val.rec != auto INFO:ImageClassificationEstimator:root.train.batch_size 128 != 8 INFO:ImageClassificationEstimator:root.train.num_workers 4 != 8 INFO:ImageClassificationEstimator:root.train.num_training_samples 1281167 != -1 INFO:ImageClassificationEstimator:root.img_cls.model resnet50_v1 != resnet18_v1b INFO:ImageClassificationEstimator:root.valid.batch_size 128 != 8 INFO:ImageClassificationEstimator:root.valid.num_workers 4 != 8 INFO:ImageClassificationEstimator:} INFO:ImageClassificationEstimator:Saved config to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_0/config.yaml INFO:ImageClassificationEstimator:Start training from [Epoch 0] INFO:ImageClassificationEstimator:Epoch[0] Batch [49] Speed: 200.960290 samples/sec accuracy=0.360000 lr=0.010000 INFO:ImageClassificationEstimator:Epoch[0] Batch [99] Speed: 247.638169 samples/sec accuracy=0.450000 lr=0.010000 INFO:ImageClassificationEstimator:[Epoch 0] training: accuracy=0.450000 INFO:ImageClassificationEstimator:[Epoch 0] speed: 219 samples/sec time cost: 5.224207 INFO:ImageClassificationEstimator:[Epoch 0] validation: top1=0.737500 top5=1.000000 INFO:ImageClassificationEstimator:[Epoch 0] Current best top-1: 0.737500 vs previous 0.000000, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_0/best_checkpoint.pkl INFO:ImageClassificationEstimator:Pickled to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_0/best_checkpoint.pkl INFO:ImageClassificationEstimator:Epoch[1] Batch [49] Speed: 229.983944 samples/sec accuracy=0.610000 lr=0.010000 INFO:ImageClassificationEstimator:Epoch[1] Batch [99] Speed: 245.688549 samples/sec accuracy=0.626250 lr=0.010000 INFO:ImageClassificationEstimator:[Epoch 1] training: accuracy=0.626250 INFO:ImageClassificationEstimator:[Epoch 1] Current best top-1: 0.790000 vs previous 0.737500, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_0/best_checkpoint.pkl INFO:ImageClassificationEstimator:[Epoch 1] speed: 234 samples/sec time cost: 4.995436 INFO:ImageClassificationEstimator:[Epoch 1] validation: top1=0.790000 top5=1.000000 INFO:ImageClassificationEstimator:Pickled to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_0/best_checkpoint.pkl INFO:ImageClassificationEstimator:modified configs( != ): { INFO:ImageClassificationEstimator:root.train.lr 0.1 != 0.01 INFO:ImageClassificationEstimator:root.train.data_dir ~/.mxnet/datasets/imagenet != auto INFO:ImageClassificationEstimator:root.train.rec_train_idx ~/.mxnet/datasets/imagenet/rec/train.idx != auto INFO:ImageClassificationEstimator:root.train.rec_train ~/.mxnet/datasets/imagenet/rec/train.rec != auto INFO:ImageClassificationEstimator:root.train.epochs 10 != 2 INFO:ImageClassificationEstimator:root.train.rec_val_idx ~/.mxnet/datasets/imagenet/rec/val.idx != auto INFO:ImageClassificationEstimator:root.train.rec_val ~/.mxnet/datasets/imagenet/rec/val.rec != auto INFO:ImageClassificationEstimator:root.train.batch_size 128 != 8 INFO:ImageClassificationEstimator:root.train.num_workers 4 != 8 INFO:ImageClassificationEstimator:root.train.num_training_samples 1281167 != -1 INFO:ImageClassificationEstimator:root.img_cls.model resnet50_v1 != mobilenetv3_small INFO:ImageClassificationEstimator:root.valid.batch_size 128 != 8 INFO:ImageClassificationEstimator:root.valid.num_workers 4 != 8 INFO:ImageClassificationEstimator:} INFO:ImageClassificationEstimator:Saved config to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_1/config.yaml INFO:ImageClassificationEstimator:Start training from [Epoch 0] INFO:ImageClassificationEstimator:Epoch[0] Batch [49] Speed: 125.261239 samples/sec accuracy=0.322500 lr=0.010000 INFO:ImageClassificationEstimator:Epoch[0] Batch [99] Speed: 138.288260 samples/sec accuracy=0.383750 lr=0.010000 INFO:ImageClassificationEstimator:[Epoch 0] training: accuracy=0.383750 INFO:ImageClassificationEstimator:[Epoch 0] Current best top-1: 0.666250 vs previous 0.000000, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_1/best_checkpoint.pkl INFO:ImageClassificationEstimator:[Epoch 0] speed: 130 samples/sec time cost: 8.767079 INFO:ImageClassificationEstimator:[Epoch 0] validation: top1=0.666250 top5=1.000000 INFO:ImageClassificationEstimator:Pickled to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_1/best_checkpoint.pkl INFO:ImageClassificationEstimator:Epoch[1] Batch [49] Speed: 126.517375 samples/sec accuracy=0.590000 lr=0.010000 INFO:ImageClassificationEstimator:Epoch[1] Batch [99] Speed: 137.455892 samples/sec accuracy=0.616250 lr=0.010000 INFO:ImageClassificationEstimator:[Epoch 1] training: accuracy=0.616250 INFO:ImageClassificationEstimator:[Epoch 1] speed: 130 samples/sec time cost: 8.718589 INFO:ImageClassificationEstimator:[Epoch 1] validation: top1=0.777500 top5=1.000000 INFO:ImageClassificationEstimator:[Epoch 1] Current best top-1: 0.777500 vs previous 0.666250, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_1/best_checkpoint.pkl INFO:ImageClassificationEstimator:Pickled to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_1/best_checkpoint.pkl INFO:ImageClassificationEstimator:Unpickled from /var/lib/jenkins/workspace/workspace/autogluon-tutorial-image-classification-v3/docs/_build/eval/tutorials/image_prediction/b5e152a0/.trial_0/best_checkpoint.pkl INFO:gluoncv.auto.tasks.image_classification:Finished, total runtime is 54.13 s INFO:gluoncv.auto.tasks.image_classification:{ 'best_config': { 'estimator': , 'gpus': [0], 'img_cls': { 'batch_norm': False, 'last_gamma': False, 'model': 'resnet18_v1b', 'use_gn': False, 'use_pretrained': True, 'use_se': False}, 'train': { 'batch_size': 8, 'crop_ratio': 0.875, 'data_dir': 'auto', 'dtype': 'float32', 'epochs': 2, 'hard_weight': 0.5, 'input_size': 224, 'label_smoothing': False, 'log_interval': 50, 'lr': 0.01, 'lr_decay': 0.1, 'lr_decay_epoch': '40, 60', 'lr_decay_period': 0, 'lr_mode': 'step', 'mixup': False, 'mixup_alpha': 0.2, 'mixup_off_epoch': 0, 'mode': '', 'momentum': 0.9, 'no_wd': False, 'num_training_samples': -1, 'num_workers': 8, 'output_lr_mult': 0.1, 'pretrained_base': True, 'rec_train': 'auto', 'rec_train_idx': 'auto', 'rec_val': 'auto', 'rec_val_idx': 'auto', 'resume_epoch': 0, 'start_epoch': 0, 'teacher': None, 'temperature': 20, 'transfer_lr_mult': 0.01, 'use_rec': False, 'warmup_epochs': 0, 'warmup_lr': 0.0, 'wd': 0.0001}, 'valid': {'batch_size': 8, 'num_workers': 8}}, 'total_time': 54.130815505981445, 'train_acc': 0.61625, 'valid_acc': 0.7775} .. parsed-literal:: :class: output Top-1 val acc: 0.777 The BO searcher can be configured by ``search_options``, see :class:`autogluon.core.searcher.GPFIFOSearcher`. Load the test dataset and evaluate: .. code:: python top1, top5 = predictor.evaluate(test_data) print('Test acc on hold-out data:', top1) .. parsed-literal:: :class: output Test acc on hold-out data: 0.75 Note that ``num_trials=2`` above is only used to speed up the tutorial. In normal practice, it is common to only use ``time_limit`` and drop ``num_trials``. Hyperband Early Stopping ~~~~~~~~~~~~~~~~~~~~~~~~ AutoGluon currently supports scheduling trials in serial order and with early stopping (e.g., if the performance of the model early within training already looks bad, the trial may be terminated early to free up resources). Here is an example of using an early stopping scheduler :class:`autogluon.core.scheduler.HyperbandScheduler`. ``scheduler_options`` is used to configure the scheduler. In this example, we run Hyperband with a single bracket, and stop/go decisions are made after 1 and 2 epochs (``grace_period``, ``grace_period * reduction_factor``): .. code:: python hyperparameters.update({ 'search_strategy': 'hyperband', 'grace_period': 1 }) The ``fit``, ``evaluate`` and ``predict`` processes are exactly the same, so we will skip training to save some time. Bayesian Optimization and Hyperband ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ While Hyperband scheduling is normally driven by a random searcher, AutoGluon also provides Hyperband together with Bayesian optimization. The tuning of expensive DL models typically works best with this combination. .. code:: python hyperparameters.update({ 'search_strategy': 'bayesopt_hyperband', 'grace_period': 1 }) For a comparison of different search algorithms and scheduling strategies, see :ref:`course_alg`. For more options using ``fit``, see :class:`autogluon.vision.ImagePredictor`.