| Name | statictorch JSON |
| Version |
0.0.4
JSON |
| download |
| home_page | None |
| Summary | Provides type annotations for torch tensors. |
| upload_time | 2025-02-22 06:17:23 |
| maintainer | None |
| docs_url | None |
| author | None |
| requires_python | None |
| license | MIT License |
| keywords |
torch
pytorch
typing
|
| VCS |
 |
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
Basic usages (only specify the count of dimensions):
```python
import torch
from statictorch import *
def f(x: Tensor2d):
pass
f(torch.zeros([2, 3])) # No runtime error, but static type checkers might say: "Tensor" is not assignable to "Tensor2d"
f(Tensor2d(torch.zeros([2, 3]))) # It's ok.
```
Note that `TensorNd()` directly return the given tensor. So at runtime you can't distinguish `x` and `TensorNd(x)`, and `isinstance(TensorNd(x), TensorNd)` will return `False`.
`TensorNd` is defined as generic classes, so that you could add a dimension descriptor for them:
```python
from statictorch import *
class Batch(TensorDimensionDescriptor):
pass
class Channel(TensorDimensionDescriptor):
pass
class Sample(TensorDimensionDescriptor):
pass
def train_on(x: Tensor3d[Batch, Channel, Sample], y: Tensor3d[Channel, Batch, Sample]):
...
def load_data() -> tuple[Tensor3d[Batch, Channel, Sample], Tensor3d[Batch, Channel, Sample]]:
...
data_x, data_y = load_data()
train_on(data_x, data_y) # Pylance: Argument of type "Tensor3d[Batch, Channel, Sample]" cannot be assigned to parameter "y" of type "Tensor3d[Channel, Batch, Sample]" in function "train_on"
# To solve the problem:
data_y_transposed = data_y.transpose(1, 0)
train_on(data_x, Tensor3d(data_y_transposed))
# Of course in some cases you want to force passing data_y, just simply cheating the type checker with:
train_on(data_x, Tensor3d(data_y))
```
`typing.cast` is also a good idea, especially when you want to call functions like `torch.stack` on a `list[TensorNd]`:
```python
import statictorch
import typing
from torch import Tensor
def work_on(tensors: list[Tensor]):
...
my_tensors: list[statictorch.Tensor0d] = []
work_on(my_tensors) # Pylance: Argument of type "list[Tensor0d]" cannot be assigned to parameter "tensors" of type "list[Tensor]" in function "work_on"
# To solve the problem:
work_on(typing.cast(list[Tensor], my_tensors))
# If you find typing.cast too long:
work_on(statictorch.anify(my_tensors))
```
Raw data
{
"_id": null,
"home_page": null,
"name": "statictorch",
"maintainer": null,
"docs_url": null,
"requires_python": null,
"maintainer_email": null,
"keywords": "torch, pytorch, typing",
"author": null,
"author_email": "yueyinqiu <yueyinqiu@outlook.com>",
"download_url": "https://files.pythonhosted.org/packages/b3/de/6d4edb346febe7044a6c8960db95536091f3346e54edeb1a31fbe4d2c8d7/statictorch-0.0.4.tar.gz",
"platform": null,
"description": "Basic usages (only specify the count of dimensions):\r\n\r\n```python\r\nimport torch\r\nfrom statictorch import *\r\n\r\ndef f(x: Tensor2d):\r\n pass\r\n\r\nf(torch.zeros([2, 3])) # No runtime error, but static type checkers might say: \"Tensor\" is not assignable to \"Tensor2d\"\r\nf(Tensor2d(torch.zeros([2, 3]))) # It's ok.\r\n```\r\n\r\nNote that `TensorNd()` directly return the given tensor. So at runtime you can't distinguish `x` and `TensorNd(x)`, and `isinstance(TensorNd(x), TensorNd)` will return `False`.\r\n\r\n`TensorNd` is defined as generic classes, so that you could add a dimension descriptor for them:\r\n\r\n```python\r\nfrom statictorch import *\r\n\r\nclass Batch(TensorDimensionDescriptor):\r\n pass\r\n\r\nclass Channel(TensorDimensionDescriptor):\r\n pass\r\n\r\nclass Sample(TensorDimensionDescriptor):\r\n pass\r\n\r\ndef train_on(x: Tensor3d[Batch, Channel, Sample], y: Tensor3d[Channel, Batch, Sample]):\r\n ...\r\n\r\ndef load_data() -> tuple[Tensor3d[Batch, Channel, Sample], Tensor3d[Batch, Channel, Sample]]:\r\n ...\r\n\r\ndata_x, data_y = load_data()\r\ntrain_on(data_x, data_y) # Pylance: Argument of type \"Tensor3d[Batch, Channel, Sample]\" cannot be assigned to parameter \"y\" of type \"Tensor3d[Channel, Batch, Sample]\" in function \"train_on\"\r\n\r\n# To solve the problem:\r\ndata_y_transposed = data_y.transpose(1, 0)\r\ntrain_on(data_x, Tensor3d(data_y_transposed))\r\n\r\n# Of course in some cases you want to force passing data_y, just simply cheating the type checker with:\r\ntrain_on(data_x, Tensor3d(data_y))\r\n```\r\n\r\n`typing.cast` is also a good idea, especially when you want to call functions like `torch.stack` on a `list[TensorNd]`:\r\n\r\n```python\r\nimport statictorch\r\nimport typing\r\nfrom torch import Tensor\r\n\r\ndef work_on(tensors: list[Tensor]):\r\n ...\r\n\r\nmy_tensors: list[statictorch.Tensor0d] = []\r\nwork_on(my_tensors) # Pylance: Argument of type \"list[Tensor0d]\" cannot be assigned to parameter \"tensors\" of type \"list[Tensor]\" in function \"work_on\"\r\n\r\n# To solve the problem:\r\nwork_on(typing.cast(list[Tensor], my_tensors))\r\n\r\n# If you find typing.cast too long:\r\nwork_on(statictorch.anify(my_tensors))\r\n```\r\n",
"bugtrack_url": null,
"license": "MIT License",
"summary": "Provides type annotations for torch tensors.",
"version": "0.0.4",
"project_urls": {
"Repository": "https://github.com/yueyinqiu/statictorch.git"
},
"split_keywords": [
"torch",
" pytorch",
" typing"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "f8c135547f3014af7bdea22bd0a1349cb0150de8e917a05efe93468477a62a19",
"md5": "a08de35ba303b74e3fb71ca51edcefef",
"sha256": "bf9ec3a5ea642c71e2133465fe6e65a4bd417dff01397e6cfdfdc14c38885914"
},
"downloads": -1,
"filename": "statictorch-0.0.4-py3-none-any.whl",
"has_sig": false,
"md5_digest": "a08de35ba303b74e3fb71ca51edcefef",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": null,
"size": 4184,
"upload_time": "2025-02-22T06:17:20",
"upload_time_iso_8601": "2025-02-22T06:17:20.653240Z",
"url": "https://files.pythonhosted.org/packages/f8/c1/35547f3014af7bdea22bd0a1349cb0150de8e917a05efe93468477a62a19/statictorch-0.0.4-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "b3de6d4edb346febe7044a6c8960db95536091f3346e54edeb1a31fbe4d2c8d7",
"md5": "f334597fc81012aed7c17f3d889b32a8",
"sha256": "8f7a054c46803ddbc80a7c86f20733defb6e1101bbed808a23a12730d5a22fd7"
},
"downloads": -1,
"filename": "statictorch-0.0.4.tar.gz",
"has_sig": false,
"md5_digest": "f334597fc81012aed7c17f3d889b32a8",
"packagetype": "sdist",
"python_version": "source",
"requires_python": null,
"size": 3440,
"upload_time": "2025-02-22T06:17:23",
"upload_time_iso_8601": "2025-02-22T06:17:23.033735Z",
"url": "https://files.pythonhosted.org/packages/b3/de/6d4edb346febe7044a6c8960db95536091f3346e54edeb1a31fbe4d2c8d7/statictorch-0.0.4.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-02-22 06:17:23",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "yueyinqiu",
"github_project": "statictorch",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"lcname": "statictorch"
}
