A pytorch template files generator, which supports multiple instances of dataset, dataloader, model, loss, optimizer and lr_scheduler.

• PyTorch Template Project
  • Requirements
  • Features
  • Folder Structure
  • Usage
    • Config file format
    • Using config files
    • Resuming from checkpoints
  • Using Multiple GPU
  • Customization
    • Custom CLI options
    • Dataset
    • Data Loader
    • Trainer
    • Model
    • Loss
    • Metrics
    • Additional logging
    • Validation data
    • Checkpoints
  • Tensorboard Visualization
  • Contribution
  • TODOs
  • License
  • Acknowledgements

• Bash (Linux)
• Python >= 3.6
• requirements.txt
• Apex

• Clear folder structure which is suitable for many deep learning projects.
• .json config file support for convenient parameter tuning.
• Support multi-dataset, multi-dataloader, multi-model, multi-optimizer, multi-loss, multi-optimizer and multi-lr_scheduler. And all of above can be constructed in .json config!
• By adding symbolic to /usr/local/bin, you can execute torch_new_project under all paths.
• Customizable command line options for more convenient parameter tuning.
• Checkpoint saving and resuming.
• Abstract base classes for faster development:
  • BaseTrainer handles checkpoint saving/resuming, training process logging, and initialize all kinds of objects.
  • BaseDataLoader handles batch generation, data shuffling, and validation data splitting.
  • BaseModel : currently not implemented.
• Additional features compared with pytorch-template:

Multiple datasets like domain adaption training will use source dataset and target dataset, so do data_loaders. Multiple models like GAN. Generator and Discriminator. Multiple losses, optimizers, lr_schedulers can be found in many ML papers.

If the paths of train/valid/test are already given, they can be directly put in the section in datasets, data_loaders.

This is the flow chart of this template:

When there are more than one module, for example,

• data_loaders/first_loader.py and data_loaders/second_loader.py
• trainers/first_trainer.py and trainers/second_trainer.py
• models/model1.py and models/model2.py

Each of them has some classes. In parse_config.py, ConfigParser.init_obj() can automatically import the specified class by using importlib.

In metric part, I add two commonly used metrics AUROC/AUPRC. These two metrics need to be computed on whole epoch, so the compute method is different from accuracy.

Continue from AUROC/AUPRC, I revise the MetricTracker, which is moved to models/metric.py. The MetricTracker can record both iteration-based metrics (iter_record) and epoch-based metrics (epoch_record).

Cross validation is supported. Class Cross_Valid in mains/cross_validation.py records the index of cross validation. The model and metric results of each fold are saved.

Also, multi-process cross validation is supported, which allows you to run many folds simultaneously. To enable multiprocessing, add flag --mp. You can decide how many processes you want to run at a time by specifying --n_jobs <int>.

⚠️ Caveat: If your dataset is large, running many processes may cost a lot of RAM. Be careful to adjust the number of processes and the number of workers in the data_loaders part of config.

RTX GPU can use apex to do automatic mixed precision training.

Use optuna to find best hyperparameters.

python3 mains/main.py -c config.json --mode train --optuna

Pytorch-Template/
│
├── base/ - abstract base classes
│
├── configs/ - configurations for training
│
├── data/ - default directory for storing input data
│
├── data_loaders/ - anything about data loading goes here
│
├── log/ - directory for storing running logs
│
├── logger/ - module for tensorboard visualization and logging
│
├── mains/ - main, train and test
│
├── models/ - models, losses, and metrics
│
├── output/ - test information
│
├── saved/ - train information
│
├── scripts/ - scripts for *.sh
│
├── trainers/ - trainers
│
├── tune/ - objectives for optuna h.p. search
│
├── utils/ - small utility functions
│
└── parse_config.py - class to handle config file and cli options

• wc

  wc -l **/*.* *.*
  

• cloc

  sudo apt install cloc
  cloc --vcs=git --by-file
  

Config file is in .json format, see configs/dataset_model.json:

There are some examples config files in configs/examples/*.json.

• MNIST dataset
• ImageNet dataset (The data need to be downloaded by yourself)
• Adult dataset

Try ./scripts/examples.sh -r to run example configs.

Modify the configurations in .json config files, then run:

python mains/main.py --config config.json --mode train

You can resume from a previously saved checkpoint by:

python mains/main.py --config config.json --mode train --resume path/to/checkpoint

You can enable multi-GPU training by setting n_gpu argument of the config file to larger number. If configured to use smaller number of gpu than available, first n devices will be used by default. Specify indices of available GPUs by cuda environmental variable.

python mains/main.py --device 2,3 -c config.json --mode train

This is equivalent to

CUDA_VISIBLE_DEVICES=2,3 python mains/main.py -c config.py --mode train

Use the torch_new_project.sh script to make your new project directory with template files.

1. Add this line to ~/.bashrc:

export pytorch_template=/path/to/pytorch-template

1. Add symbolic link at /usr/local/bin so that you can run this script everywhere.

sudo ln -s $pytorch_template/scripts/new_project/torch_new_project.sh /usr/local/bin/torch_new_project

1. torch_new_project ProjectName produces a new project folder named 'ProjectName' will be made. This script will filter out unneccessary files listed in copy_exclude.

Changing values of config file is a clean, safe and easy way of tuning hyperparameters. However, sometimes it is better to have command line options if some values need to be changed too often or quickly.

This template uses the configurations stored in the json file by default, but by registering custom options as follows you can change some of them using CLI flags.

# simple class-like object having 3 attributes, `flags`, `type`, `target`.
CustomArgs = collections.namedtuple('CustomArgs', 'flags type target')
options = [
    CustomArgs(['--lr', '--learning_rate'], type=float, target="optimizer;args;lr"),
    CustomArgs(['--bs', '--batch_size'], type=int, target="data_loader;args;batch_size")
    # options added here can be modified by command line flags.
]

target argument should be sequence of keys, which are used to access that option in the config dict. In this example, target for the learning rate option is "optimizer;args;lr" because config['optimizer']['args']['lr'] points to the learning rate. python mains/main.py -c config.json --mode train --bs 256 runs training with options given in config.json except for the batch size which is increased to 256 by command line options.

• Writing your own data loader

1. Inherit BaseDataLoader

   BaseDataLoader is a subclass of torch.utils.data.DataLoader, you can use either of them.

   BaseDataLoader handles:

   • Generating next batch
   • Data shuffling
   • Generating validation data loader by calling BaseDataLoader.split_validation()

• DataLoader Usage

  BaseDataLoader is an iterator, to iterate through batches:

  for batch_idx, (x_batch, y_batch) in data_loader:
      pass

• Example

  Please refer to data_loaders/examples/MNIST_loader.py for an MNIST data loading example.

• Writing your own trainer

1. Inherit BaseTrainer

   BaseTrainer handles:

   • Training process logging
   • Checkpoint saving
   • Checkpoint resuming
   • Reconfigurable performance monitoring for saving current best model, and early stop training.
     • If config monitor is set to max val_accuracy, which means then the trainer will save a checkpoint model_best.pth when validation accuracy of epoch replaces current maximum.
     • If config early_stop is set, training will be automatically terminated when model performance does not improve for given number of epochs. This feature can be turned off by passing 0 to the early_stop option, or just deleting the line of config.

2. Implementing abstract methods

   You need to implement _train_epoch() for your training process, if you need validation then you can implement _valid_epoch() as in trainers/trainer.py

• Example

  Please refer to trainers/trainer.py for MNIST training.

• Iteration-based training

  Trainer.__init__ takes an optional argument, len_epoch which controls number of batches(steps) in each epoch.

• Writing your own model

1. Inherit BaseModel

   BaseModel handles:

   • Inherited from torch.nn.Module
   • __str__: Modify native print function to prints the number of trainable parameters.

2. Implementing abstract methods

   Implement the foward pass method forward()

• Example

  Please refer to models/examples/LeNet.py for a LeNet example.

Custom loss functions can be implemented in 'model/loss.py'. Use them by changing the name given in "loss" in config file, to corresponding name.

Metric functions are located in models/metric.py.

You can monitor multiple metrics by providing a list in the configuration file, e.g.:

"metrics": {
    "per_iteration": ["accuracy", "top_k_acc"],
    "per_epoch": ["AUROC", "AUPRC"],
    "pick_threshold": {
        "is_ftn": true,
        "type": "Youden_J",
        "kwargs": {
            "beta": 1.0
        }
    }
}

If you have additional information to be logged, in _train_epoch() of your trainer class, merge them with log as shown below before returning:

additional_log = {"gradient_norm": g, "sensitivity": s}
log.update(additional_log)
return log

You can test trained model by running mains/test.py passing path to the trained checkpoint by --resume argument.

To split validation data from a data loader, call BaseDataLoader._train_valid_split(), then it will return a data loader for validation of size specified in your config file. The validation_split can be a ratio of validation set per total data(0.0 <= float < 1.0), or the number of samples (0 <= int < n_total_samples).

Note: the _train_valid_split() method will modify the original data loader Note: _train_valid_split() will return None if "validation_split" is set to 0

You can specify the name of the training session in config files:

name: MNIST_LeNet,

The checkpoints will be saved in saved/name/run_id/model/checkpoint_epoch_n, with timestamp in mmdd_HHMMSS format.

A copy of config file will be saved in the same folder.

Note: checkpoints contain:

{
    'epoch': epoch,
    'models': {key: value.state_dict() for key, value in self.models.items()},
    'optimizers': {key: value.state_dict() for key, value in self.optimizers.items()},
    'monitor_best': self.mnt_best,
}

This template supports Tensorboard visualization by using either torch.utils.tensorboard or TensorboardX.

1. Install

   If you are using pytorch 1.1 or higher, install tensorboard by pip install tensorboard>=1.14.0.

   Otherwise, you should install tensorboardx. Follow installation guide in TensorboardX.

2. Run training

   Make sure that tensorboard option in the config file is turned on.

    "tensorboard" : true
   

3. Open Tensorboard server

   Type tensorboard --logdir saved/EXP/run_id/log/ at the project root, then server will open at http:https://localhost:6006

By default, values of loss and metrics specified in config file, input images, and histogram of model parameters will be logged. If you need more visualizations, use add_scalar('tag', data), add_image('tag', image), etc in the trainer._train_epoch method. add_something() methods in this template are basically wrappers for those of tensorboardX.SummaryWriter and torch.utils.tensorboard.SummaryWriter modules.

Note: You don't have to specify current steps, since WriterTensorboard class defined at logger/visualization.py will track current steps.

Feel free to contribute any kind of function or enhancement, here the codes using black formatter

Code should use the black to format the codes before committing.

• Revise errors in trainer/examples and test_examples/
• Support more tensorboard functions

This project is forked and enhanced from the project pytorch-template