pip install https://github.com/jaeminSon/torchfetch
>>> import torchfetch
# get dataloader (classification)
>>> kwargs_dataloader = {"num_workers": 16, "pin_memory": True, "batch_size": 32, "shuffle": True}
>>> dataloader = torchfetch.get_dataloader(data="Test/objects/data/image_csv",
preprocess="Test/objects/preprocess/imagenet.json",
augment="Test/objects/augment/imagenet.json",
**kwargs_dataloader)
>>> for batch in dataloader:
... <process batch>
# get dataloader (detection)
>>> kwargs_dataloader = {"num_workers": 16, "pin_memory": True, "batch_size": 16, "shuffle": True, "collate_fn": lambda x: x}
>>> dataloader = torchfetch.get_dataloader(data="Test/objects/data/detection1",
preprocess="Test/objects/preprocess/detection1.json",
augment="Test/objects/augment/cocodetection.json",
**kwargs_dataloader)
# public network class instantiation
>>> network = torchfetch.instantiate_network("resnet50")
>>> network = torchfetch.instantiate_network("resnet34", **{"pretrained":True})
# load a network from model name (or recipe name for private architecture)
>>> network = torchfetch.get_pretrained_network("Test/objects/recipe/private_arch.json")
# network class instantiation
>>> network = torchfetch.instantiate_network("Test/objects/recipe/private_arch.json")
# get checkpoint
>>> checkpoint = torchfetch.get_checkpoint("Test/objects/recipe/private_arch.json")
# get network state dict
>>> model_state_dict = torchfetch.get_model_state_dict("Test/objects/recipe/private_arch.json")
# get optimizer state dict
>>> optimizer_state_dict = torchfetch.get_optimizer_state_dict("Test/objects/recipe/private_arch.json")
# install graphviz and pydeps
linux: $ sudo apt install graphviz
mac: $ brew install graphviz
$ pip install pydeps
# draw dependency graph
$ pydeps torchfetch --reverse --only torchfetch --exclude-exact torchfetch
# No cycle found with the following command
$ pydeps torchfetch --reverse --only torchfetch --exclude-exact torchfetch --show-cycles
