aprint

A Python print package for those who don't care the actual elements of array-like data (e.g. numpy.array, torch.tensor) within a complex data structure.

Background

By default, print will display the actual elements in array-like data, such as numpy.array or torch.tensor, which is not very friendly for those who don't care about the actual elements. Inspired by the leading scientific computing language Julia, this package will print the type and shape of the data structure, as well as the type and shape of the elements within the data structure.

Installation

pip install aprint

For the latest version, you can install from github:

pip install git+https://github.com/huangyxi/aprint.git

Usage

>>> import numpy as np
>>> import torch
>>> from aprint import aprint

>>> aprint(np.array([1, 2, 3]))
int64[3] numpy.ndarray

>>> aprint({'a': torch.zeros(16, 32).cuda(), 2: [np.array([1, 2]), (torch.zeros(2).to_sparse(), torch.zeros(2,3).to_sparse_csr(), torch.zeros(2,3,4).to_sparse_csc()), {'5', 6, 7.}]}, indent_str='| ')
{ # dict with len=2
| 'a': float32[16×32] torch.Tensor at 'cuda:0' with grad=False,
| 2: [ # list with len=3
| | int64[2] numpy.ndarray,
| | ( # tuple with len=3
| | | float32[2] torch.Tensor at 'cpu' with layout='COO', grad=False,
| | | float32[2×3] torch.Tensor at 'cpu' with layout='CSR', grad=False,
| | | float32[2×3×4] torch.Tensor at 'cpu' with layout='CSC', grad=False,
| | ),
| | { # set with len=3
| | | '5',
| | | 6,
| | | 7.0,
| | },
| ],
}

>>> aprint([1, [2, [3, [4, [5]]]]], max_depth=3, indent_str="⋮ ") # default max_depth=5
[ # list with len=2
⋮ 1,
⋮ [ # list with len=2
⋮ ⋮ 2,
⋮ ⋮ [ # list with len=2
⋮ ⋮ ⋮ 3,
⋮ ⋮ ⋮ [ # list with len=2 ... ],
⋮ ⋮ ],
⋮ ],
]

>>> torch.tensor([1, 2]) | aprint # pipeline print (only for torch.Tensor and built-in iterators)
int64[2] torch.Tensor at 'cpu' with grad=False

Roadmap

• pipeline support for torch.Tensor
• continuous integration
• support class inheritance
  • support iterators' classes
  • support objects' classes
• support more types
  • torch.nn.parameter.Parameter
• colorful output
• glance of the data

License

MIT License