Torch Autodiff Utility
Project description
Torch Autodiff Utility
Compatibility: |
|
Availability: |
|
Status: |
|
This library is a collection of utility functions that are used in PyTorch (re-)implementations of projects from the Grimme group. In particular, the tad-mctc library provides:
-
autograd functions (Jacobian, Hessian)
-
atomic data (radii, EN, example molecules, ...)
-
batch utility (packing, masks, ...)
-
conversion functions (numpy, atomic symbols/numbers, ...)
-
coordination numbers (DFT-D3, DFT-D4, EEQ)
-
io (reading/writing coordinate files)
-
molecular properties (bond lengths/orders/angles, moment of inertia, ...)
-
safeops (autograd-safe implementations of common functions)
-
typing (base class for tensor-like behavior of arbitrary classes)
-
units
The name is inspired by the Fortran pendant "modular computation tool chain library" (mctc-lib).
Installation
pip
tad-mctc can easily be installed with pip
.
pip install tad-mctc
conda
tad-mctc is also available from conda
.
conda install tad-mctc
From source
This project is hosted on GitHub at tad-mctc/tad-mctc. Obtain the source by cloning the repository with
git clone https://github.com/tad-mctc/tad-mctc
cd tad-mctc
We recommend using a conda environment to install the package. You can setup the environment manager using a mambaforge installer. Install the required dependencies from the conda-forge channel.
mamba env create -n torch -f environment.yaml
mamba activate torch
Install this project with pip
in the environment
pip install .
The following dependencies are required
- numpy
- opt_einsum
- psutil
- pytest (tests only)
- torch
Development
For development, additionally install the following tools in your environment.
mamba install black covdefaults mypy pre-commit pylint pytest pytest-cov pytest-xdist tox
pip install pytest-random-order
With pip, add the option -e
for installing in development mode, and add [dev]
for the development dependencies
pip install -e .[dev]
The pre-commit hooks are initialized by running the following command in the root of the repository.
pre-commit install
For testing all Python environments, simply run tox
.
tox
Note that this randomizes the order of tests but skips "large" tests. To modify this behavior, tox
has to skip the optional posargs.
tox -- test
Examples
The following example shows how to calculate the coordination number used in the EEQ model for a single structure.
import torch
import tad_mctc as mctc
numbers = mctc.convert.symbol_to_number(symbols="C C C C N C S H H H H H".split())
# coordinates in Bohr
positions = torch.tensor(
[
[-2.56745685564671, -0.02509985979910, 0.00000000000000],
[-1.39177582455797, +2.27696188880014, 0.00000000000000],
[+1.27784995624894, +2.45107479759386, 0.00000000000000],
[+2.62801937615793, +0.25927727028120, 0.00000000000000],
[+1.41097033661123, -1.99890996077412, 0.00000000000000],
[-1.17186102298849, -2.34220576284180, 0.00000000000000],
[-2.39505990368378, -5.22635838332362, 0.00000000000000],
[+2.41961980455457, -3.62158019253045, 0.00000000000000],
[-2.51744374846065, +3.98181713686746, 0.00000000000000],
[+2.24269048384775, +4.24389473203647, 0.00000000000000],
[+4.66488984573956, +0.17907568006409, 0.00000000000000],
[-4.60044244782237, -0.17794734637413, 0.00000000000000],
]
)
# calculate EEQ coordination number
cn = mctc.ncoord.cn_eeq(numbers, positions)
torch.set_printoptions(precision=10)
print(cn)
# tensor([3.0519218445, 3.0177774429, 3.0132560730, 3.0197706223,
# 3.0779352188, 3.0095663071, 1.0991339684, 0.9968624115,
# 0.9943327904, 0.9947233200, 0.9945874214, 0.9945726395])
The next example shows the calculation of the coordination number used in DFT-D4 for a batch of structures.
import torch
import tad_mctc as mctc
# S22 system 4: formamide dimer
numbers = mctc.batch.pack((
mctc.convert.symbol_to_number("C C N N H H H H H H O O".split()),
mctc.convert.symbol_to_number("C O N H H H".split()),
))
# coordinates in Bohr
positions = mctc.batch.pack((
torch.tensor([
[-3.81469488143921, +0.09993441402912, 0.00000000000000],
[+3.81469488143921, -0.09993441402912, 0.00000000000000],
[-2.66030049324036, -2.15898251533508, 0.00000000000000],
[+2.66030049324036, +2.15898251533508, 0.00000000000000],
[-0.73178529739380, -2.28237795829773, 0.00000000000000],
[-5.89039325714111, -0.02589114569128, 0.00000000000000],
[-3.71254944801331, -3.73605775833130, 0.00000000000000],
[+3.71254944801331, +3.73605775833130, 0.00000000000000],
[+0.73178529739380, +2.28237795829773, 0.00000000000000],
[+5.89039325714111, +0.02589114569128, 0.00000000000000],
[-2.74426102638245, +2.16115570068359, 0.00000000000000],
[+2.74426102638245, -2.16115570068359, 0.00000000000000],
]),
torch.tensor([
[-0.55569743203406, +1.09030425468557, 0.00000000000000],
[+0.51473634678469, +3.15152550263611, 0.00000000000000],
[+0.59869690244446, -1.16861263789477, 0.00000000000000],
[-0.45355203669134, -2.74568780438064, 0.00000000000000],
[+2.52721209544999, -1.29200800956867, 0.00000000000000],
[-2.63139587595376, +0.96447869452240, 0.00000000000000],
]),
))
# calculate coordination number
cn = mctc.ncoord.cn_d4(numbers, positions)
torch.set_printoptions(precision=10)
print(cn)
# tensor([[2.6886456013, 2.6886456013, 2.6314170361, 2.6314167976,
# 0.8594539165, 0.9231414795, 0.8605306745, 0.8605306745,
# 0.8594539165, 0.9231414795, 0.8568341732, 0.8568341732],
# [2.6886456013, 0.8568335176, 2.6314167976, 0.8605306745,
# 0.8594532013, 0.9231414795, 0.0000000000, 0.0000000000,
# 0.0000000000, 0.0000000000, 0.0000000000, 0.0000000000]])
Contributing
This is a volunteer open source projects and contributions are always welcome. Please, take a moment to read the contributing guidelines.
License
This project is licensed under the Apache License, Version 2.0 (the "License"); you may not use this project's files except in compliance with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.