Using Optuna to Optimize PyTorch Ignite Hyperparameters
PyTorch Ignite + Optuna!
Optuna is a hyperparameter optimization framework applicable to machine learning frameworks and black-box optimization solvers. PyTorch Ignite is a high-level library for PyTorch that helps you write compact, but full-featured, code in less lines. Combining the two of them allows for automatic tuning of hyperparameters to find the best performing models.
Creating the Objective Function
Optuna is a black-box optimizer, which means it needs an objectivefunction, which returns a numerical value to evaluate the performance of the hyperparameters, and decide where to sample in upcoming trials.
In our example, we will be doing this for identifying MNIST characters from the Optuna GitHub examples folder. In this case, the objective function starts like this:
Notice that the objective function is passed an Optuna specific argument of trial. This object is passed to the objective function to be used to specify which hyperparameters should be tuned. This returns the accuracy of the model as evaluator.state.metrics[“accuracy”], which is used by Optuna as feedback on the performance of the trial.
Defining the hyperparameters to be tuned
Optuna allows you to define the kinds and ranges of hyperparameters you want to tune directly within your code using the trial object. This saves the effort of learning specialized syntax for hyperparameters, and also means you can use normal Python code for looping through or defining your hyperparameters.
Optuna supports a variety of hyperparameter settings, which can be used to optimize floats, integers, or discrete categorical values. Numerical values can be suggested from a logarithmic continuum as well. In our MNIST example, we optimize the hyperparameters here:
The dropout percentage is defined by trial.suggest_uniform(“dropout_rate”, 0, 1), which gives a float value between 0 and 1.
The number of input layers to be tuned is given from fc2_input_dim = trial.suggest_int(“fc2_input_dim”, 40, 80), which gives an integer value from 40 to 80, which will be labelled in Optuna as fc2_input_dim.
While not used in this example, for hyperparameters that should vary by orders of magnitude, such as learning rates, use something like trial.suggest_loguniform('learning_rate', 1e-5, 1000), which will vary the values from .00001 to 0.1. Also, categorical selection from a list is possible with trial.suggest_categorical(‘optimizer’, [‘SGD’, ‘Adam’]).
Running the Trials
The default sampler in Optuna Tree-structured Parzen Estimater (TPE), which is a form of Bayesian Optimization. Optuna uses TPE to search more efficiently than a random search, by choosing points closer to previous good results.
To run the trials, create a study object, which sets the direction of optimization ("maximize" or "minimize"), along with other settings. Then, the study object run with optimize(objective, n_trials=100, timeout=600), to do one hundred trials, with a timeout of ten minutes.
Each trial is chosen after evaluating all the trials that have been previously done, using a sampler to make a smart guess where the best values hyperparameters can be found. Optuna provides Tree-structured Parzen Estimator (TPE) samplers, which is a kind of bayesian optimization, as the default sampler.
The best values from the trials can be accessed through study.best_trial, and other methods of viewing the trials, such as formatting in a dataframe, are available.
Pruning — Early Stopping of Poor Trials
Pruning trials is a form of early-stopping which terminates unpromising trials, so that computing time can be used for trials that show more potential. In order to do pruning, it’s necessary to open up the black-box of the Objective function some more to provide intermittent feedback on how the trial is going to Optuna, so it can compare the progress with the progress of other trials, and decide whether to stop the trial early, and provide a method to receive a method from Optuna when the trial should be terminated, and also allow the trial in session to terminate cleanly after recording the results. Fortunately, Optuna provides an integration for PyTorch Ignite (PyTorchIgnitePruningHandler) pruning that provides all of these functions.
After importing the PyTorchIgnitePruningHandler, passing it as a early_stop_callback to the trainer allows Ignite to do the pruning. The monitor argument of the PyTorchIgnitePruningHandler function references the PyTorch Ignite model.state dictionary , such as accuracy or loss.
To the Future, and Beyond!
For those interested, Optuna has many other features, including a visualizations, alternative samplers, optimizers, and pruning algorithms, as well as the ability to create user-defined versions as well. If you have more computing resources available, Optuna provides an easy interface for parallel trials to increase tuning speed.
Give Optuna a try!
This post uses pytorch-ignite v0.3.0and optuna v1.1.0.
