# MDFS (MultiDimensional Feature Selection) for Python
MDFS is a library to assist in MultiDimensional Feature Selection (MDFS),
i.e. feature selection that accounts for multidimensional interactions
in the dataset. To learn more about MDFS, please visit the
[MDFS website][mdfs-web].
This project is the implementation of the MDFS library for Python.
Functionality-wise, it is aligned with the
[R version of the MDFS library][mdfs-r], but the interface differs to
make it more native to the Python ecosystem (i.e. _pythonic_) and to
free it from early assumptions carried on for backward compatibility in R.
## License
This software is released the same as the R MDFS library: under the
[GNU General Public License (GPL) v3][gpl-3].
## Copyright
The copyrights are held by Radosław Piliszek (the package maintainer
and author), Abraham Kaczmarski (major contributor to the new interface),
Krzysztof Mnich and Witold Rudnicki (authors of the MDFS method).
## Changelog
See [the common changelog][changelog].
## Library structure
The library consists of a single package module: `mdfs`, which exports
all the user-facing functionality.
## Introduction for beginners
The `mdfs` package module needs to be imported. Then, the main function
to run is, aptly named, `run`. It accepts a numpy data matrix data and
its corresponding decision, and returns a dictionary with the details of
analysis, including the entry for `relevant_variables` which gives the
indices of variables deemed relevant under chosen conditions.
## Interface differences between R and Python
### Function names
The following list gives the translation between R functions and their
Python counterparts.
- `MDFS` = `run`
- `ComputeMaxInfoGains` = `compute_max_ig`
- `ComputeInterestingTuples` = `compute_tuples`
- `ComputePValue` = `fit_p_value`
- `Discretize` = `discretize`
- `GetRange` = `get_suggested_range`
- `GenContrastVariables` = `gen_contrast_variables`
### Function parameter names
Function parameter names have been adjusted to avoid the dot (`.`),
replacing it with an underscore (`_`).
### No global seed in Python
There is no global seed in use. All functions depending on PRNG take
a `seed` parameter.
### Quirks
Due to the way the Python-C interface is implemented in this library with
`numpy` views, there is one quirk to be aware of. Functions returning
a `Structure` subclass object do so without incurring a copy. Properties
present on such objects return views, not copies. These views do not protect
the result from being garbage collected (i.e., think of them as weak
references to the underlying data). Thus, to avoid freed memory reads,
keep the original structures around when using these views or copy
data elsewhere as necessary.
This quirk might be lifted in the future.
[mdfs-web]: https://www.mdfs.it/
[mdfs-r]: https://cran.r-project.org/package=MDFS
[gpl-3]: https://www.gnu.org/licenses/gpl-3.0.en.html
[changelog]: https://www.mdfs.it/CHANGELOG
Raw data
{
"_id": null,
"home_page": "",
"name": "mdfs",
"maintainer": "",
"docs_url": null,
"requires_python": "",
"maintainer_email": "Rados\u0142aw Piliszek <r.piliszek@uwb.edu.pl>",
"keywords": "feature selection,classification,machine learning,information theory",
"author": "",
"author_email": "Rados\u0142aw Piliszek <r.piliszek@uwb.edu.pl>",
"download_url": "https://files.pythonhosted.org/packages/e3/84/2d0a30c9526307f21ba651df194cb8c76db6de36a106fd429889332a3a03/mdfs-1.5.2.tar.gz",
"platform": null,
"description": "# MDFS (MultiDimensional Feature Selection) for Python\n\nMDFS is a library to assist in MultiDimensional Feature Selection (MDFS),\ni.e. feature selection that accounts for multidimensional interactions\nin the dataset. To learn more about MDFS, please visit the\n[MDFS website][mdfs-web].\n\nThis project is the implementation of the MDFS library for Python.\nFunctionality-wise, it is aligned with the\n[R version of the MDFS library][mdfs-r], but the interface differs to\nmake it more native to the Python ecosystem (i.e. _pythonic_) and to\nfree it from early assumptions carried on for backward compatibility in R.\n\n## License\n\nThis software is released the same as the R MDFS library: under the\n[GNU General Public License (GPL) v3][gpl-3].\n\n## Copyright\n\nThe copyrights are held by Rados\u0142aw Piliszek (the package maintainer\nand author), Abraham Kaczmarski (major contributor to the new interface),\nKrzysztof Mnich and Witold Rudnicki (authors of the MDFS method).\n\n## Changelog\n\nSee [the common changelog][changelog].\n\n## Library structure\n\nThe library consists of a single package module: `mdfs`, which exports\nall the user-facing functionality.\n\n## Introduction for beginners\n\nThe `mdfs` package module needs to be imported. Then, the main function\nto run is, aptly named, `run`. It accepts a numpy data matrix data and\nits corresponding decision, and returns a dictionary with the details of\nanalysis, including the entry for `relevant_variables` which gives the\nindices of variables deemed relevant under chosen conditions.\n\n## Interface differences between R and Python\n\n### Function names\n\nThe following list gives the translation between R functions and their\nPython counterparts.\n\n- `MDFS` = `run`\n- `ComputeMaxInfoGains` = `compute_max_ig`\n- `ComputeInterestingTuples` = `compute_tuples`\n- `ComputePValue` = `fit_p_value`\n- `Discretize` = `discretize`\n- `GetRange` = `get_suggested_range`\n- `GenContrastVariables` = `gen_contrast_variables`\n\n### Function parameter names\n\nFunction parameter names have been adjusted to avoid the dot (`.`),\nreplacing it with an underscore (`_`).\n\n### No global seed in Python\n\nThere is no global seed in use. All functions depending on PRNG take\na `seed` parameter.\n\n### Quirks\n\nDue to the way the Python-C interface is implemented in this library with\n`numpy` views, there is one quirk to be aware of. Functions returning\na `Structure` subclass object do so without incurring a copy. Properties\npresent on such objects return views, not copies. These views do not protect\nthe result from being garbage collected (i.e., think of them as weak\nreferences to the underlying data). Thus, to avoid freed memory reads,\nkeep the original structures around when using these views or copy\ndata elsewhere as necessary.\nThis quirk might be lifted in the future.\n\n\n[mdfs-web]: https://www.mdfs.it/\n[mdfs-r]: https://cran.r-project.org/package=MDFS\n[gpl-3]: https://www.gnu.org/licenses/gpl-3.0.en.html\n[changelog]: https://www.mdfs.it/CHANGELOG\n",
"bugtrack_url": null,
"license": "",
"summary": "A library for MultiDimensional Feature Selection (MDFS)",
"version": "1.5.2",
"split_keywords": [
"feature selection",
"classification",
"machine learning",
"information theory"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "e3842d0a30c9526307f21ba651df194cb8c76db6de36a106fd429889332a3a03",
"md5": "a3c0a4f23b89488d174af34c3c208dd5",
"sha256": "e1a25b44471a0a567b6cb9e90d660cbe1e354618100d4a08b340f996858855ec"
},
"downloads": -1,
"filename": "mdfs-1.5.2.tar.gz",
"has_sig": false,
"md5_digest": "a3c0a4f23b89488d174af34c3c208dd5",
"packagetype": "sdist",
"python_version": "source",
"requires_python": null,
"size": 24147,
"upload_time": "2023-04-19T08:13:54",
"upload_time_iso_8601": "2023-04-19T08:13:54.978325Z",
"url": "https://files.pythonhosted.org/packages/e3/84/2d0a30c9526307f21ba651df194cb8c76db6de36a106fd429889332a3a03/mdfs-1.5.2.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2023-04-19 08:13:54",
"github": false,
"gitlab": false,
"bitbucket": false,
"lcname": "mdfs"
}
JSON