torchstruct 0.1.5.post0

No project description provided

torchstruct

Wrap your multiple torch.Tensors into single TensorStruct and use it like you are using torch.Tensor.

Installation

pip install torchstruct

Testing

PYTHONPATH=. pytest

Examples

import torch
from torchstruct import TensorStruct

# Initialization
ts = TensorStruct.zeros({
    'obs': (2,),
    'rew': (1,),
    'done': (1,)
}, prefix_shape=(10,), dtype=torch.float32, device='cpu')

raw_data = {
    'obs': torch.randn((10, 2)),
    'rew': torch.randn((10, 1)),
    'done': torch.randn((10, 1))
}

# Assigning
ts[:] = raw_data

# Indexing
ts[2:4]
ts['rew']

# Calling PyTorch methods
ts.unsqueeze(dim=0)
ts.sum(dim=0)