MLMBench - MachineLearning Molecular Benchmarch
MLMBench collects datasets and splits them to do FAIR ML benchmarks. MLMBench can be used with different ML algorithms and data representations for molecular property/activity predictions and more.
The scope of this code is:
- keep a simple API representation
- no need of other libraries
- keep the dataset offline and represented as CSV file (RFC 4180 standard) or SMILES string list.
Splits are made using well-known rational approaches such as:
- random split
- meaningful split for model target extrapolation
- meaningful split for chemical diversity extrapolation
- literature published split
The datasets are stored in the "data" directory in subfolders. Every subfolder needs the following files with the following names:
- Readme.txt: explain some dataset info (provenience, type of data, descriptors version, and so on)
- cv.splits: the split required to do a fair trainin, test, validation in any ml algorithm
- dataset.csv: the matrix of features
- target.csv: the matrix of target/targets
- dataset.smi: the smiles list
Install
pip3 install mlmbench
Split types per dataset
mlmbench includes for every dataset two different splits:
- random split using "mkrndsplits.py" starting from a list of names
- target extrapolation using "mktgtextrapsplits.py" starting from the target file. In this case, the algorithm will first import the target file, and then for every column, rank from min to max the queue and split the ordered target into "N" splits selected by the user. This split aims to check for "extrapolation."
- literature split (if available). In this case, we try to preserve particular splits published by users.
Available datasets
- BACE-moleculenet
- BACE-random
- BACE-tgt_extrapolation
- FU-random
- FU-tgt_extrapolation
- HLMCLint-random
- HLMCLint-tgt_extrapolation
- MeltingPoint-random
- MeltingPoint-tgt_extrapolation
- NIR_Gasoline-random
- NIR_Gasoline-tgt_extrapolation
- SteroidsLSS-isomers
- SteroidsLSS-random
- SteroidsLSS-tgt_extrapolation
- esol-chemdiversity
- esol-random
- esol-tgt_extrapolation
- logDpH7.4-random
- logDpH7.4-tgt_extrapolation
How to use
#!/usr/bin/env python3
from mlmbench.data import Datasets
ds = Datasets()
print(ds.get_available_datasets())
print(f'Dataset info: {ds.get_info("esol-random")}')
for train_data, test_data, val_data in ds.ttv_generator("esol-random"):
print("train ", train_data["xdata"].shape, train_data["target"].shape, len(train_data["smi"]))
print("test ", test_data["xdata"].shape, test_data["target"].shape, len(test_data["smi"]))
print("val ", val_data["xdata"].shape, val_data["target"].shape, len(val_data["smi"]))
# Do ml training/test validation, collect the results and store it in your
# appropriate format to do your analysis.
print("-"*40)
Submit new dataset
-
Fork the project!
-
Clone the forked project
-
Add the dataset in this form: dataset.csv: tabular data for any kind of descriptors target.csv: tabular data for one or multiple targets dataset.smi: smiles of the molecule in its appropriate format "c1ccccc1 benzene" cv.split: The split you like. This specific file needs to be compatible with the following standard. The file comprises lines representing the model, groups split by the ";" character, and every group representing the compound name, and every name is split using the "," character. i.e. train keys test keys validation keys line 1 mol1,mol2,mol3,.. ; mol200,mol201,... ; mol400,mol401,... line 2 ... line 3 ..
Readme.md: Info regarding the dataset(i.e. source and so on)
-
Create a pull request and 99.9% will be merged
Metadata
Release files for mlmbench 1.0.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| mlmbench-1.0.3.tar.gz | 49.9 MB | Without provenance | Details for mlmbench-1.0.3.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| mlmbench-1.0.3-py3-none-any.whl | Python 3 | none | any | Details for mlmbench-1.0.3-py3-none-any.whl |
Total release size: 100.5 MB
Release files / mlmbench-1.0.3.tar.gz
| Download URL | mlmbench-1.0.3.tar.gz |
|---|---|
| Size | 49.9 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
a5e2f086eadfaf422a9a753aab9fa2ba1bf1911843b3b8e26117f803362e05e5
|
|
BLAKE2b-256 checksum How to use checksums |
90d061191d6c855a8dab77c38c0457d98daef392e476cdb24c5ffd616269af36
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/3.3.0 pkginfo/1.4.2 requests/2.25.1 setuptools/52.0.0 requests-toolbelt/0.9.1 tqdm/4.57.0 CPython/3.9.2
|
Release files / mlmbench-1.0.3-py3-none-any.whl
| Download URL | mlmbench-1.0.3-py3-none-any.whl |
|---|---|
| Size | 50.6 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
820282dbe1ca12ae66a4656edddf77f8b729297fb4e4821559225e590256624e
|
|
BLAKE2b-256 checksum How to use checksums |
34c6e5f669d78199e6b0ce4c20ee9fc1d866a8683d2d8aa23e2c75a2768ccce2
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/3.3.0 pkginfo/1.4.2 requests/2.25.1 setuptools/52.0.0 requests-toolbelt/0.9.1 tqdm/4.57.0 CPython/3.9.2
|
Security for mlmbench 1.0.3
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.