Skip to content
PyTorch deep learning projects made easy.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
base remove python2 style usage of super() May 14, 2019
data_loader remove python2 style usage of super() May 14, 2019
logger change layout for TB logging, add `steps_per_sec` Apr 23, 2019
model remove python2 style usage of super() May 14, 2019
trainer remove python2 style usage of super() May 14, 2019
utils change layout for TB logging, add `steps_per_sec` Apr 23, 2019
.flake8 Add .flake8 for F401, F403 and make max-line-length = 120 Jun 14, 2018
.gitignore add items on .gitignore Sep 16, 2018
LICENSE Create LICENSE Apr 5, 2018
README.md update README.md Apr 23, 2019
config.json revert config.json option name: trainer Apr 18, 2019
new_project.py check for current directory Apr 24, 2019
parse_config.py fix typo May 8, 2019
test.py fix issue #51 about test.py May 20, 2019
train.py

README.md

PyTorch Template Project

PyTorch deep learning project made easy.

Requirements

  • Python >= 3.5
  • PyTorch >= 0.4
  • tqdm (Optional for test.py)
  • tensorboard >= 1.7.0 (Optional for TensorboardX)
  • tensorboardX >= 1.2 (Optional for TensorboardX)

Features

  • Clear folder structure which is suitable for many deep learning projects.
  • .json config file support for convenient parameter tuning.
  • 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 more.
    • BaseDataLoader handles batch generation, data shuffling, and validation data splitting.
    • BaseModel provides basic model summary.

Folder Structure

pytorch-template/
│
├── train.py - main script to start training
├── test.py - evaluation of trained model
│
├── config.json - holds configuration for training
├── parse_config.py - class to handle config file and cli options
│
├── new_project.py - initialize new project with template files
│
├── base/ - abstract base classes
│   ├── base_data_loader.py
│   ├── base_model.py
│   └── base_trainer.py
│
├── data_loader/ - anything about data loading goes here
│   └── data_loaders.py
│
├── data/ - default directory for storing input data
│
├── model/ - models, losses, and metrics
│   ├── model.py
│   ├── metric.py
│   └── loss.py
│
├── saved/
│   ├── models/ - trained models are saved here
│   └── log/ - default logdir for tensorboardX and logging output
│
├── trainer/ - trainers
│   └── trainer.py
│
├── logger/ - module for tensorboardX visualization and logging
│   ├── visualization.py
│   ├── logger.py
│   └── logger_config.json
│  
└── utils/ - small utility functions
    ├── util.py
    └── ...

Usage

The code in this repo is an MNIST example of the template. Try python train.py -c config.json to run code.

Config file format

Config files are in .json format:

{
  "name": "Mnist_LeNet",        // training session name
  "n_gpu": 1,                   // number of GPUs to use for training.
  
  "arch": {
    "type": "MnistModel",       // name of model architecture to train
    "args": {

    }                
  },
  "data_loader": {
    "type": "MnistDataLoader",         // selecting data loader
    "args":{
      "data_dir": "data/",             // dataset path
      "batch_size": 64,                // batch size
      "shuffle": true,                 // shuffle training data before splitting
      "validation_split": 0.1          // size of validation dataset. float(portion) or int(number of samples)
      "num_workers": 2,                // number of cpu processes to be used for data loading
    }
  },
  "optimizer": {
    "type": "Adam",
    "args":{
      "lr": 0.001,                     // learning rate
      "weight_decay": 0,               // (optional) weight decay
      "amsgrad": true
    }
  },
  "loss": "nll_loss",                  // loss
  "metrics": [
    "my_metric", "my_metric2"          // list of metrics to evaluate
  ],                         
  "lr_scheduler": {
    "type": "StepLR",                   // learning rate scheduler
    "args":{
      "step_size": 50,          
      "gamma": 0.1
    }
  },
  "trainer": {
    "epochs": 100,                     // number of training epochs
    "save_dir": "saved/",              // checkpoints are saved in save_dir/models/name
    "save_freq": 1,                    // save checkpoints every save_freq epochs
    "verbosity": 2,                    // 0: quiet, 1: per epoch, 2: full
  
    "monitor": "min val_loss"          // mode and metric for model performance monitoring. set 'off' to disable.
    "early_stop": 10	                 // number of epochs to wait before early stop. set 0 to disable.
  
    "tensorboardX": true,              // enable tensorboardX visualization
  }
}

Add addional configurations if you need.

Using config files

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

python train.py --config config.json

Resuming from checkpoints

You can resume from a previously saved checkpoint by:

python train.py --resume path/to/checkpoint

Using Multiple GPU

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 train.py --device 2,3 -c config.json

This is equivalent to

CUDA_VISIBLE_DEVICES=2,3 python train.py -c config.py

Customization

Project initialization

Use the new_project.py script to make your new project directory with template files. python new_project.py ../NewProject then a new project folder named 'NewProject' will be made. This script will filter out unneccessary files like cache, git files or readme file.

Custom CLI options

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 train.py -c config.json --bs 256 runs training with options given in config.json except for the batch size which is increased to 256 by command line options.

Data Loader

  • 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_loader/data_loaders.py for an MNIST data loading example.

Trainer

  • 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 trainer/trainer.py

  • Example

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

Model

  • 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 model/model.py for a LeNet example.

Loss

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.

Metrics

Metric functions are located in 'model/metric.py'.

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

"metrics": ["my_metric", "my_metric2"],

Additional logging

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 = {**log, **additional_log}
return log

Testing

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

Validation data

To split validation data from a data loader, call BaseDataLoader.split_validation(), 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 split_validation() method will modify the original data loader Note: split_validation() will return None if "validation_split" is set to 0

Checkpoints

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

"name": "MNIST_LeNet",

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

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

Note: checkpoints contain:

{
  'arch': arch,
  'epoch': epoch,
  'state_dict': self.model.state_dict(),
  'optimizer': self.optimizer.state_dict(),
  'monitor_best': self.mnt_best,
  'config': self.config
}

TensorboardX Visualization

This template supports TensorboardX visualization.

  • TensorboardX Usage
  1. Install

    Follow installation guide in TensorboardX.

  2. Run training

    Set tensorboardX option in config file true.

  3. Open tensorboard server

    Type tensorboard --logdir saved/log/ at the project root, then server will open at http://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 module.

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

Contributing

Feel free to contribute any kind of function or enhancement, here the coding style follows PEP8

Code should pass the Flake8 check before committing.

TODOs

  • Iteration-based training (instead of epoch-based)
  • Multiple optimizers
  • Configurable logging layout, checkpoint naming
  • visdom logger support
  • tensorboardX logger support
  • Adding command line option for fine-tuning
  • Multi-GPU support
  • Update the example to PyTorch 0.4
  • Learning rate scheduler
  • Deprecate BaseDataLoader, use torch.utils.data instesad
  • Load settings from config files

License

This project is licensed under the MIT License. See LICENSE for more details

Acknowledgments

This project is inspired by the project Tensorflow-Project-Template by Mahmoud Gemy

You can’t perform that action at this time.