<img src="https://github.com/chython/chytorch/assets/2013465/9e97e99b-060d-4beb-bd8c-80c895ac9b81" width="100" height="100"> Chytorch [kʌɪtɔːrtʃ]
====================
Library for modeling molecules and reactions in torch way.
Installation
------------
Use `pip install chytorch` to install release version.
Or `pip install .` in source code directory to install DEV version.
Pretrained models
-----------------
Chytorch main package doesn't include models zoo.
Each model has its own named package and can be installed separately.
Installed models can be imported as `from chytorch.zoo.<model_name> import Model`.
Usage
-----
`chytorch.nn.MoleculeEncoder` and `chytorch.nn.ReactionEncoder` - core graphormer layers for molecules and reactions.
API is combination of `torch.nn.TransformerEncoderLayer` with `torch.nn.TransformerEncoder`.
**Batch preparation:**
`chytorch.utils.data.MoleculeDataset` and `chytorch.utils.data.ReactionDataset` - Map-like on-the-fly dataset generators for molecules and reactions.
Supported `chython.MoleculeContainer` and `chython.ReactionContainer` objects, and bytes-packed structures.
`chytorch.utils.data.collate_molecules` and `chytorch.utils.data.collate_reactions` - collate functions for `torch.utils.data.DataLoader`.
Note: torch DataLoader automatically do proper collation since 1.13 release.
Example:
from chytorch.utils.data import MoleculeDataset, SMILESDataset
from torch.utils.data import DataLoader
data = ['CCO', 'CC=O']
ds = MoleculeDataset(SMILESDataset(data, cache={}))
dl = DataLoader(ds, batch_size=10)
**Forward call:**
Molecules coded as tensors of:
* atoms numbers shifted by 2 (e.g. hydrogen = 3).
0 - reserved for padding, 1 - reserved for CLS token, 2 - extra reservation.
* neighbors count, including implicit hydrogens shifted by 2 (e.g. CO = CH3OH = [6, 4]).
0 - reserved for padding, 1 - extra reservation, 2 - no-neighbors, 3 - one neighbor.
* topological distances' matrix shifted by 2 with upper limit.
0 - reserved for padding, 1 - reserved for not-connected graph components coding, 2 - self-loop, 3 - connected atoms.
Reactions coded in similar way. Molecules atoms and neighbors matrices just stacked. Distance matrices stacked on diagonal.
Reactions include additional tensor with reaction role codes for each token.
0 - padding, 1 - reaction CLS, 2 - reactants, 3 - products.
from chytorch.nn import MoleculeEncoder
encoder = MoleculeEncoder()
for b in dl:
encoder(b)
**Combine molecules and labels:**
`chytorch.utils.data.chained_collate` - helper for combining different data parts. Useful for tricky input.
from torch import stack
from torch.utils.data import DataLoader, TensorDataset
from chytorch.utils.data import chained_collate, collate_molecules, MoleculeDataset
dl = DataLoader(TensorDataset(MoleculeDataset(molecules_list), properties_tensor),
collate_fn=chained_collate(collate_molecules, stack))
**Scheduler:**
`chytorch.optim.lr_scheduler.WarmUpCosine` - Linear warmup followed with cosine-function for 0-pi range rescaled to lr_rate - decrease_coef * lr_rate interval.
**Voting NN with single hidden layer:**
`chytorch.nn.VotingClassifier`, `chytorch.nn.BinaryVotingClassifier` and `chytorch.nn.VotingRegressor` - speed optimized multiple heads for ensemble predictions.
**Helper Modules:**
`chytorch.nn.Slicer` - do tensor slicing. Useful for transformer's CLS token extraction in `torch.nn.Sequence`.
**Data Wrappers:**
In `chytorch.utils.data` module stored different data wrappers for simplifying ML workflows.
All wrappers have `torch.utils.data.Dataset` interface.
* `SizedList` - list wrapper with `size()` method. Useful with `torch.utils.data.TensorDataset`.
* `SMILESDataset` - on-the-fly smiles to `chython.MoleculeContainer` or `chython.ReactionContainer` parser.
* `LMDBMapper` - LMDB KV storage to dataset mapper.
* `PostgresMapper` - Postgres DB table to dataset mapper.
* `SMILESTokenizerDataset` - on-the-fly generator of tokenized SMILES.
* `TensorUnpack`, `StructUnpack`, `PickleUnpack` - bytes to tensor/object unpackers
Publications
------------
[1](https://doi.org/10.1021/acs.jcim.2c00344) Bidirectional Graphormer for Reactivity Understanding: Neural Network Trained to Reaction Atom-to-Atom Mapping Task
Raw data
{
"_id": null,
"home_page": "https://github.com/chython/chytorch",
"name": "chytorch",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.8,<3.12",
"maintainer_email": "",
"keywords": "",
"author": "Ramil Nugmanov",
"author_email": "nougmanoff@protonmail.com",
"download_url": "",
"platform": null,
"description": "<img src=\"https://github.com/chython/chytorch/assets/2013465/9e97e99b-060d-4beb-bd8c-80c895ac9b81\" width=\"100\" height=\"100\"> Chytorch [k\u028c\u026at\u0254\u02d0rt\u0283]\n====================\n\nLibrary for modeling molecules and reactions in torch way.\n\nInstallation\n------------\n\nUse `pip install chytorch` to install release version.\n\nOr `pip install .` in source code directory to install DEV version.\n\nPretrained models\n-----------------\n\nChytorch main package doesn't include models zoo.\nEach model has its own named package and can be installed separately.\nInstalled models can be imported as `from chytorch.zoo.<model_name> import Model`.\n\n\nUsage\n-----\n\n`chytorch.nn.MoleculeEncoder` and `chytorch.nn.ReactionEncoder` - core graphormer layers for molecules and reactions.\nAPI is combination of `torch.nn.TransformerEncoderLayer` with `torch.nn.TransformerEncoder`. \n\n**Batch preparation:**\n\n`chytorch.utils.data.MoleculeDataset` and `chytorch.utils.data.ReactionDataset` - Map-like on-the-fly dataset generators for molecules and reactions.\nSupported `chython.MoleculeContainer` and `chython.ReactionContainer` objects, and bytes-packed structures.\n\n`chytorch.utils.data.collate_molecules` and `chytorch.utils.data.collate_reactions` - collate functions for `torch.utils.data.DataLoader`.\n\nNote: torch DataLoader automatically do proper collation since 1.13 release.\n\nExample:\n\n from chytorch.utils.data import MoleculeDataset, SMILESDataset\n from torch.utils.data import DataLoader\n\n data = ['CCO', 'CC=O']\n ds = MoleculeDataset(SMILESDataset(data, cache={}))\n dl = DataLoader(ds, batch_size=10)\n\n**Forward call:**\n\nMolecules coded as tensors of:\n* atoms numbers shifted by 2 (e.g. hydrogen = 3).\n 0 - reserved for padding, 1 - reserved for CLS token, 2 - extra reservation.\n* neighbors count, including implicit hydrogens shifted by 2 (e.g. CO = CH3OH = [6, 4]).\n 0 - reserved for padding, 1 - extra reservation, 2 - no-neighbors, 3 - one neighbor.\n* topological distances' matrix shifted by 2 with upper limit.\n 0 - reserved for padding, 1 - reserved for not-connected graph components coding, 2 - self-loop, 3 - connected atoms.\n\nReactions coded in similar way. Molecules atoms and neighbors matrices just stacked. Distance matrices stacked on diagonal.\nReactions include additional tensor with reaction role codes for each token.\n0 - padding, 1 - reaction CLS, 2 - reactants, 3 - products.\n\n from chytorch.nn import MoleculeEncoder\n \n encoder = MoleculeEncoder()\n for b in dl:\n encoder(b)\n\n**Combine molecules and labels:**\n\n`chytorch.utils.data.chained_collate` - helper for combining different data parts. Useful for tricky input.\n\n from torch import stack\n from torch.utils.data import DataLoader, TensorDataset\n from chytorch.utils.data import chained_collate, collate_molecules, MoleculeDataset\n\n dl = DataLoader(TensorDataset(MoleculeDataset(molecules_list), properties_tensor),\n collate_fn=chained_collate(collate_molecules, stack))\n\n\n**Scheduler:**\n\n`chytorch.optim.lr_scheduler.WarmUpCosine` - Linear warmup followed with cosine-function for 0-pi range rescaled to lr_rate - decrease_coef * lr_rate interval.\n\n**Voting NN with single hidden layer:**\n\n`chytorch.nn.VotingClassifier`, `chytorch.nn.BinaryVotingClassifier` and `chytorch.nn.VotingRegressor` - speed optimized multiple heads for ensemble predictions.\n\n**Helper Modules:**\n\n`chytorch.nn.Slicer` - do tensor slicing. Useful for transformer's CLS token extraction in `torch.nn.Sequence`.\n\n**Data Wrappers:**\n\nIn `chytorch.utils.data` module stored different data wrappers for simplifying ML workflows.\nAll wrappers have `torch.utils.data.Dataset` interface.\n\n* `SizedList` - list wrapper with `size()` method. Useful with `torch.utils.data.TensorDataset`. \n* `SMILESDataset` - on-the-fly smiles to `chython.MoleculeContainer` or `chython.ReactionContainer` parser.\n* `LMDBMapper` - LMDB KV storage to dataset mapper.\n* `PostgresMapper` - Postgres DB table to dataset mapper.\n* `SMILESTokenizerDataset` - on-the-fly generator of tokenized SMILES.\n* `TensorUnpack`, `StructUnpack`, `PickleUnpack` - bytes to tensor/object unpackers\n\n\nPublications\n------------\n\n[1](https://doi.org/10.1021/acs.jcim.2c00344) Bidirectional Graphormer for Reactivity Understanding: Neural Network Trained to Reaction Atom-to-Atom Mapping Task\n\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "Library for modeling molecules and reactions in torch way",
"version": "1.62",
"project_urls": {
"Homepage": "https://github.com/chython/chytorch"
},
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "15af6cb6636c21189731b6e817b80f81fa05dd974972ab1d0fd46de1e1344ecc",
"md5": "da77cde96fea712365cce866f150d19c",
"sha256": "4444f7d0d43246d11f436873b911a135208bde2d1a3c26957f01d13055eafbde"
},
"downloads": -1,
"filename": "chytorch-1.62-cp310-cp310-macosx_12_0_x86_64.whl",
"has_sig": false,
"md5_digest": "da77cde96fea712365cce866f150d19c",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.8,<3.12",
"size": 173575,
"upload_time": "2024-03-01T12:35:12",
"upload_time_iso_8601": "2024-03-01T12:35:12.162689Z",
"url": "https://files.pythonhosted.org/packages/15/af/6cb6636c21189731b6e817b80f81fa05dd974972ab1d0fd46de1e1344ecc/chytorch-1.62-cp310-cp310-macosx_12_0_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "5d429859094f61ad3d35af0d421c157e763daf15d4d1fc6b404524390452cfe3",
"md5": "bea7219a5d434f3351d3f438df696948",
"sha256": "0726bbc613d2bbd1b0b89f344b8a7c3595996ed75b257bff831b88ff77980beb"
},
"downloads": -1,
"filename": "chytorch-1.62-cp310-cp310-manylinux_2_31_x86_64.whl",
"has_sig": false,
"md5_digest": "bea7219a5d434f3351d3f438df696948",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.8,<3.12",
"size": 547717,
"upload_time": "2024-03-01T12:35:02",
"upload_time_iso_8601": "2024-03-01T12:35:02.335507Z",
"url": "https://files.pythonhosted.org/packages/5d/42/9859094f61ad3d35af0d421c157e763daf15d4d1fc6b404524390452cfe3/chytorch-1.62-cp310-cp310-manylinux_2_31_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "d67e4f3e8bd9865b8243ad76cce8bf9c963f7dfc3fc906f856d89f3237a6c3fb",
"md5": "edc53c42476488726d1aeecc6a3f5029",
"sha256": "e6f8d529e398b7a0f3177666a6348368579bc0c46d91cb7972981d349e75965e"
},
"downloads": -1,
"filename": "chytorch-1.62-cp310-cp310-win_amd64.whl",
"has_sig": false,
"md5_digest": "edc53c42476488726d1aeecc6a3f5029",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.8,<3.12",
"size": 164137,
"upload_time": "2024-03-01T12:36:22",
"upload_time_iso_8601": "2024-03-01T12:36:22.358092Z",
"url": "https://files.pythonhosted.org/packages/d6/7e/4f3e8bd9865b8243ad76cce8bf9c963f7dfc3fc906f856d89f3237a6c3fb/chytorch-1.62-cp310-cp310-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "90729d076ff9559d872cee151b97ee9a5881fb0a89776c5f6f5a713fe7ae9cc1",
"md5": "dadd5352436ab48a277945806e1d112e",
"sha256": "dd635dbc4df516e338d99c11d0e12f9b691fbb4537373b9ba759676cac8f48cb"
},
"downloads": -1,
"filename": "chytorch-1.62-cp311-cp311-macosx_12_0_x86_64.whl",
"has_sig": false,
"md5_digest": "dadd5352436ab48a277945806e1d112e",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.8,<3.12",
"size": 249957,
"upload_time": "2024-03-01T12:35:12",
"upload_time_iso_8601": "2024-03-01T12:35:12.970558Z",
"url": "https://files.pythonhosted.org/packages/90/72/9d076ff9559d872cee151b97ee9a5881fb0a89776c5f6f5a713fe7ae9cc1/chytorch-1.62-cp311-cp311-macosx_12_0_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "760817018e0c300e2d4d80744388bb787664720f21aee785ee5f02f393460f3a",
"md5": "1ce9a36994e1b7a6e791e662d30bffb3",
"sha256": "5269e9f6895be6db86b02f6c4e93b17dbbffaab7b4f8aa5183a60a07f63e5b3f"
},
"downloads": -1,
"filename": "chytorch-1.62-cp311-cp311-manylinux_2_31_x86_64.whl",
"has_sig": false,
"md5_digest": "1ce9a36994e1b7a6e791e662d30bffb3",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.8,<3.12",
"size": 583651,
"upload_time": "2024-03-01T12:35:01",
"upload_time_iso_8601": "2024-03-01T12:35:01.572075Z",
"url": "https://files.pythonhosted.org/packages/76/08/17018e0c300e2d4d80744388bb787664720f21aee785ee5f02f393460f3a/chytorch-1.62-cp311-cp311-manylinux_2_31_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "c7b61c0e366037720b766e4aeb4b8135c4253bc5af1829582c5c0ada94404d63",
"md5": "340a2dc9f8e0c487ddfe26123d7a5d97",
"sha256": "78f9e804a5661012de2b5d546ae03de47d2afc7513cf181fbeb9eea593aa9a9a"
},
"downloads": -1,
"filename": "chytorch-1.62-cp311-cp311-win_amd64.whl",
"has_sig": false,
"md5_digest": "340a2dc9f8e0c487ddfe26123d7a5d97",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.8,<3.12",
"size": 163973,
"upload_time": "2024-03-01T12:36:06",
"upload_time_iso_8601": "2024-03-01T12:36:06.542767Z",
"url": "https://files.pythonhosted.org/packages/c7/b6/1c0e366037720b766e4aeb4b8135c4253bc5af1829582c5c0ada94404d63/chytorch-1.62-cp311-cp311-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "125d0d62f6ec2fbe5a0036860eb25820f72e64ff8f5d8ff9d15590b1e657c5af",
"md5": "28c9234b2bd8c49ffd9a502dc2b62fae",
"sha256": "ba1415d69c716d023aa08f26568314a7882617ad5c52c3156df2559ae22c2547"
},
"downloads": -1,
"filename": "chytorch-1.62-cp38-cp38-macosx_12_0_x86_64.whl",
"has_sig": false,
"md5_digest": "28c9234b2bd8c49ffd9a502dc2b62fae",
"packagetype": "bdist_wheel",
"python_version": "cp38",
"requires_python": ">=3.8,<3.12",
"size": 173480,
"upload_time": "2024-03-01T12:35:14",
"upload_time_iso_8601": "2024-03-01T12:35:14.738305Z",
"url": "https://files.pythonhosted.org/packages/12/5d/0d62f6ec2fbe5a0036860eb25820f72e64ff8f5d8ff9d15590b1e657c5af/chytorch-1.62-cp38-cp38-macosx_12_0_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "54165702157587108c7302bee48a18de75e53dc0d627a8c86db535f95268801f",
"md5": "a7471852caf6c34c5a0819fd81089c33",
"sha256": "91c3a64b4eccb8c59934d21a26741f79550b03df7775d1cf80129376c69efcf9"
},
"downloads": -1,
"filename": "chytorch-1.62-cp38-cp38-manylinux_2_31_x86_64.whl",
"has_sig": false,
"md5_digest": "a7471852caf6c34c5a0819fd81089c33",
"packagetype": "bdist_wheel",
"python_version": "cp38",
"requires_python": ">=3.8,<3.12",
"size": 564342,
"upload_time": "2024-03-01T12:34:58",
"upload_time_iso_8601": "2024-03-01T12:34:58.477494Z",
"url": "https://files.pythonhosted.org/packages/54/16/5702157587108c7302bee48a18de75e53dc0d627a8c86db535f95268801f/chytorch-1.62-cp38-cp38-manylinux_2_31_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "d23bd14fde65feac8ec2337e485207596d18cf74adcdd5401e6d418925dda444",
"md5": "12b4971f4c880357cb9b46fd73b9c223",
"sha256": "b3d6dfbc1e244a9148c3af05faf552066971cd7501c226564e47f11823c893c5"
},
"downloads": -1,
"filename": "chytorch-1.62-cp38-cp38-win_amd64.whl",
"has_sig": false,
"md5_digest": "12b4971f4c880357cb9b46fd73b9c223",
"packagetype": "bdist_wheel",
"python_version": "cp38",
"requires_python": ">=3.8,<3.12",
"size": 164827,
"upload_time": "2024-03-01T12:36:08",
"upload_time_iso_8601": "2024-03-01T12:36:08.545451Z",
"url": "https://files.pythonhosted.org/packages/d2/3b/d14fde65feac8ec2337e485207596d18cf74adcdd5401e6d418925dda444/chytorch-1.62-cp38-cp38-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "7f0a286752e1a213e26deb2bff8c516070b58f07a4acb8b34e61f3be2b55c4a2",
"md5": "83b41394f57c1fd7a0d5a92634280c2a",
"sha256": "549dd76d6d49017fba9dbdc81ffc0d2d7614cf7328f1d837d1a0df996b7ff9d4"
},
"downloads": -1,
"filename": "chytorch-1.62-cp39-cp39-macosx_12_0_x86_64.whl",
"has_sig": false,
"md5_digest": "83b41394f57c1fd7a0d5a92634280c2a",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.8,<3.12",
"size": 174211,
"upload_time": "2024-03-01T12:35:11",
"upload_time_iso_8601": "2024-03-01T12:35:11.318605Z",
"url": "https://files.pythonhosted.org/packages/7f/0a/286752e1a213e26deb2bff8c516070b58f07a4acb8b34e61f3be2b55c4a2/chytorch-1.62-cp39-cp39-macosx_12_0_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "ccba214daa8f3b924b9ca99d0c0671ca6ba9483fc689f6515680b0d6ee7aadc9",
"md5": "7f426635951d70600704542c4e01207e",
"sha256": "5d7a698db9f6a58c149cb2c8f990bc57e743caab3cbb46152f6902dda6e642a6"
},
"downloads": -1,
"filename": "chytorch-1.62-cp39-cp39-manylinux_2_31_x86_64.whl",
"has_sig": false,
"md5_digest": "7f426635951d70600704542c4e01207e",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.8,<3.12",
"size": 550771,
"upload_time": "2024-03-01T12:35:05",
"upload_time_iso_8601": "2024-03-01T12:35:05.339645Z",
"url": "https://files.pythonhosted.org/packages/cc/ba/214daa8f3b924b9ca99d0c0671ca6ba9483fc689f6515680b0d6ee7aadc9/chytorch-1.62-cp39-cp39-manylinux_2_31_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "4dc16753434be5c31bb8822c7bbc1324a5d71bad3a32a4ef19b0a3aec273220e",
"md5": "7e447e7a51679284f7e58c60abd39c5f",
"sha256": "169fc3910e835e2c87f790569ebf58fcb4a46165b1cddbc36b28a18c306cf5be"
},
"downloads": -1,
"filename": "chytorch-1.62-cp39-cp39-win_amd64.whl",
"has_sig": false,
"md5_digest": "7e447e7a51679284f7e58c60abd39c5f",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.8,<3.12",
"size": 164711,
"upload_time": "2024-03-01T12:35:54",
"upload_time_iso_8601": "2024-03-01T12:35:54.419596Z",
"url": "https://files.pythonhosted.org/packages/4d/c1/6753434be5c31bb8822c7bbc1324a5d71bad3a32a4ef19b0a3aec273220e/chytorch-1.62-cp39-cp39-win_amd64.whl",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-03-01 12:35:12",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "chython",
"github_project": "chytorch",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "chytorch"
}
JSON
