kjappelbaum / oximachine_featurizer

featurizing MOFs for the oximachine

oximachine_featurizer

Mine oxidation states for structures from the (MOF) subset of the CSD and calculate features for them. Runscripts are automatically installed for the most important steps. Some of these runscripts contain hard coded paths, that would need to be updated. This code generates inputs that can be used with the learnmofox package to replicate our work [1].

If you're just interested in using a pre-trained model, the oximachinerunner package.

⚠️ Warning: For the mining of the oxidation states, you need the CSD Python API. You need to export the CSD_HOME path. Due to the licensing issues, this cannot be done automatically.

Installation

The commands below automatically install several command-line tools (CLI) which are detailed below.

The full process should take some seconds.

Latest version

To install the latest version of the software with all dependencies, you can use

pip install git+https://github.com/kjappelbaum/oximachine_featurizer.git

Stable release

pip install oximachine_featurizer

How to use it

To run the default featurization on one structure you can use the CLI

run_featurization <structure> <outdir>

for each metal center this should take seconds if there is no disorder.

Some output can be found on the MaterialsCloud Archive (doi: 10.24435/materialscloud:2019.0085/v1 ).

More details can be found in the documentation.

Example usage

The use of the main functions of this package is shown in the Jupyter Notebook in the example directory. It contains some example structures and the output, which should be produces in seconds.

Testing the installation

For testing, you can---as it is done for the continuous integration (CI)---use pytest and run the files in the test directory. For example

pip install pytest
pytest test/main

References

[1] Jablonka, Kevin Maik; Ongari, Daniele; Moosavi, Seyed Mohamad; Smit, Berend (2020): Using Collective Knowledge to Assign Oxidation States. ChemRxiv. Preprint. https://doi.org/10.26434/chemrxiv.11604129.v1

oximachine_featurizer's Issues

add issue and PR template

Use black instead of yapf

add release please

Add coverage badge

how to download CSD entry name with oxidation states?

In paper, you mentioned using CSD API. but not sure how it looks like.
can u give the code to retrieve the MOF oxidation dataset?

Test with which python versions it works

do I really want to stick with GPL?

Remove prospector

Check for the number of atoms

add more test for the ordering of the metals

to make sure that the ordering is preserved

package 'test' installed

you're trying to avoid this here
https://github.com/kjappelbaum/oximachine_featurizer/blob/master/setup.py#L63

but since the folder is called test , it slips through

$ pip uninstall oximachine-featurizer
Found existing installation: oximachine-featurizer 0.2.14
Uninstalling oximachine-featurizer-0.2.14:
  Would remove:
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/bin/run_featurization
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/bin/run_mine_mp
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/bin/run_parsing
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/bin/run_parsing_reference
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/lib/python3.8/site-packages/oximachine_featurizer-0.2.14.dist-info/*
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/lib/python3.8/site-packages/oximachine_featurizer/*
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/lib/python3.8/site-packages/run/*
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/lib/python3.8/site-packages/test/*
  Would not remove (might be manually added):
    /Users/leopold/Applications/miniconda3/envs/aiida-lsmo-2/lib/python3.8/site-packages/test/test_oximachine.py

Set up readthedocs

matminer dependency now broken due to pymatgen update

Steps to reproduce the problem

noticed in aiida-lsmo https://github.com/lsmo-epfl/aiida-lsmo/runs/2459918160

/opt/hostedtoolcache/Python/3.7.10/x64/lib/python3.7/importlib/__init__.py:127: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
tests/test_oxidation_state.py:11: in <module>
    from aiida_lsmo.calcfunctions.oxidation_state import compute_oxidation_states
aiida_lsmo/calcfunctions/oxidation_state.py:5: in <module>
    import oximachinerunner
/opt/hostedtoolcache/Python/3.7.10/x64/lib/python3.7/site-packages/oximachinerunner/__init__.py:15: in <module>
    from oximachine_featurizer import featurize
/opt/hostedtoolcache/Python/3.7.10/x64/lib/python3.7/site-packages/oximachine_featurizer/__init__.py:9: in <module>
    from .featurize import FeatureCollector, GetFeatures, featurize
/opt/hostedtoolcache/Python/3.7.10/x64/lib/python3.7/site-packages/oximachine_featurizer/featurize.py:16: in <module>
    from matminer.featurizers.site import CrystalNNFingerprint, GaussianSymmFunc
/opt/hostedtoolcache/Python/3.7.10/x64/lib/python3.7/site-packages/matminer/featurizers/site.py:8: in <module>
    from matminer.utils.data import MagpieData
/opt/hostedtoolcache/Python/3.7.10/x64/lib/python3.7/site-packages/matminer/utils/data.py:17: in <module>
    from pymatgen import Element
E   ImportError: cannot import name 'Element' from 'pymatgen' (unknown location)

I guess you'll want to update the matminer dependency to a version that isn't broken (or restrict the pymatgen version <2021)

Dependabot couldn't authenticate with https://pypi.python.org/simple/

Dependabot couldn't authenticate with https://pypi.python.org/simple/.

You can provide authentication details in your Dependabot dashboard by clicking into the account menu (in the top right) and selecting 'Config variables'.

View the update logs.

Move the exclusion list to separate module and better indicate why they are excluded

Currently there is one long list in the class and I started mixing up structures that are excluded because they are special test structures, due to EDA or due to errors found in some previous iterations of the model. We should clean this in separate lists that we import from a separate module. Generally, in the longer run, this part of the code also needs to be fixed as it is really stupid and slow to loop over all this list for each structure. One can easily improve it quite a lot using e.g. sets.

concurrent featurization

the issue is a bit that we're CPU bound and usually it is probably easier to parallelize over structures and not sites