This is a package designed to provide a single utility: showing you the types!
If you frequently end up handling lists of tuples of tensors, lists of tensors, nested lists and tuples, or other similarly convoluted structures, SMTT might be able to make your life just a bit easier. Instead of thinking about needing len, .shape, .size(), or printing out the structure and parsing the brackets and parens, you can just do tt(data), and get an instant overview of the structure.
from showmethetypes import SMTT
import torch
tt = SMTT()
structure = [(torch.randn(2,3), torch.randn(500, 5, 128)) for _ in range(5)]
tt(structure)list (5)
└── tuple
├── Tensor (dtype: torch.float32)
│ │ (device: cpu)
│ ├── dim_0 (2)
│ └── dim_1 (3)
└── Tensor (dtype: torch.float32)
│ (device: cpu)
├── dim_0 (500)
├── dim_1 (5)
└── dim_2 (128)
Done locally through pip
git clone https://github.com/InfiniteSwerve/ShowMeTheTypes.git
cd ShowMeTheTypes
pip install -e .
SMTT will by default display the type for whatever you pass it, and traverse all nested types if the passed object has a __getitem__ method.
The default behavior is to iterate across all subtypes, which means it will print out everything the object contains. Some types, like lists, are usually monomorphic, so SMTT will assume monomorphism and only print out the nested structure in the first item of the list.
SMTT will also infer your current imports and use the specific handlers available for any libraries in your locals().
SMTT is quite fresh, and so the current coverage only includes objects that I've needed to work with directly. If you feel sad about any missing types, feel free to raise an issue requesting it, or even make a PR! Creating new handlers for types is quite straightforward.
You can run the tests with pytest --nbval.
Since the core functionality doesn't change, the testing focus is on regression tests. For this we use pytest and nbval, using the cell outputs of jupyter notebooks to make sure the untouched cell outputs haven't changed.
The libraries with any coverage are listed here, and you can find detailed coverage information in handlers.py.
| Library | Total Coverage |
|---|---|
| stdlib | No |
| Pytorch | No |
| Numpy | No |