lightningdata-modules 0.1.9

Pre-packages Pytorch-Lightning datasets

LightningData Modules

Pre-packages Pytorch-Lightning datasets.

Installation

To install this library, simply run the following command:

pip install lightingdata_modules

Installing the LightningData Modules should automatically install suitable dependencies.

Customized Dataset Support

LightningData Modules downloads, installs and prepares customized Pytorch Lightning datamodules with just one line of code.

Example from Domain Adaptation (DomainNet):

# import the custom DomainNet module
import lightningdata_modules.domain_adaptation.domainNet_datamodule as domainNet

# initialize the data module with the "real" domain from DomainNet
my_data_module = domainNet.DomainNetDataModule(data_dir="./dataset/", domain="real")

The data module can now be passed to the Pytorch Lighting Trainer instance or be used as a standalone Dataloader:

# check if the dataset already exists in the data_dir, download domain-specific lmdb database from cloud storage otherwise
my_data_module.prepare_data()

# initialize the train and test set according to setup properties
my_data_module.setup()

# create the train DataLoader on-the-fly
train_loader = my_data_module.train_dataloader()

List of Available Datasets

Domain Adaptation	Federated Learning	Meta Learning
DomainNet	EMNIST	Omiglot
Office31		Mini Imagenet
OfficeHome
Digit-Five

Class Diagram

Ack

Thanks to TorchMeta: https://github.com/tristandeleu/pytorch-meta