{
    "info": {
        "author": "Thibault Cl\u00e9rice",
        "author_email": "",
        "bugtrack_url": null,
        "classifiers": [
            "License :: OSI Approved :: MIT License",
            "Programming Language :: Python",
            "Programming Language :: Python :: 3",
            "Programming Language :: Python :: 3.6",
            "Topic :: Text Processing :: Linguistic"
        ],
        "description": "\n# Le Boucher d'Amsterdam\n\nBoudams, or \"Le boucher d'Amsterdam\", is a deep-learning tool built for tokenizing Latin or Medieval French languages.\n\n## How to cite\n\nAn article has been published about this work : https://hal.archives-ouvertes.fr/hal-02154122v1\n\n```text\n@unpublished{clerice:hal-02154122,\n  TITLE = {{Evaluating Deep Learning Methods for Word Segmentation of Scripta Continua Texts in Old French and Latin}},\n  AUTHOR = {Cl{\\'e}rice, Thibault},\n  URL = {https://hal.archives-ouvertes.fr/hal-02154122},\n  NOTE = {working paper or preprint},\n  YEAR = {2019},\n  MONTH = Jun,\n  KEYWORDS = {convolutional network ; scripta continua ; tokenization ; Old French ; word segmentation},\n  PDF = {https://hal.archives-ouvertes.fr/hal-02154122/file/Evaluating_Deep_Learning_Methods_for_Tokenization_of_Scripta_Continua_in_Old_French_and_Latin%284%29.pdf},\n  HAL_ID = {hal-02154122},\n  HAL_VERSION = {v1},\n}\n\n```\n\n## How to\n\nInstall the usual way you install python stuff: `python setup.py install` (**Python >= 3.6**)).\n\nThe config file can be kickstarted using `boudams template config.json`, we recommend using the following settings :\n\n- `linear-conv-no-pos` for the model, as it is not limited by the input size;\n- `normalize` and `lower` to `True` depending on your dataset size.\n\nThe initial dataset is pretty small but if you want to build with your own, it's fairly simple : you need data in the \nfollowing shape : `\"samesentence<TAB>same sentence\"` where the first element is the same than the second but with no\nspace and they are separated by tabs (`\\t`, marked here as `<TAB>`).\n\n\n```json\n{\n    \"name\": \"model\",\n    \"max_sentence_size\": 150,\n    \"network\": {\n        \"emb_enc_dim\": 256,\n        \"enc_n_layers\": 10,\n        \"enc_kernel_size\": 3,\n        \"enc_dropout\": 0.25\n    },\n    \"model\": \"linear-conv-no-pos\",\n    \"learner\": {\n        \"lr_grace_periode\": 2,\n        \"lr_patience\": 2,\n        \"lr\": 0.0001\n    },\n    \"label_encoder\": {\n        \"normalize\": true,\n        \"lower\": true\n    },\n    \"datasets\": {\n        \"test\": \"./test.tsv\",\n        \"train\": \"./train.tsv\",\n        \"dev\": \"./dev.tsv\",\n        \"random\": true\n    }\n}\n```\n\nThe best architecture I find for medieval French was Conv to Linear without POS using the following setup:\n\n```json\n{\n    \"network\": {\n        \"emb_enc_dim\": 256,\n        \"enc_n_layers\": 10,\n        \"enc_kernel_size\": 5,\n        \"enc_dropout\": 0.25\n    },\n    \"model\": \"linear-conv-no-pos\",\n    \"batch_size\": 64,\n    \"learner\": {\n        \"lr_grace_periode\": 2,\n        \"lr_patience\": 2,\n        \"lr\": 0.00005,\n        \"lr_factor\": 0.5\n    }\n}\n```\n\n\n## Credits\n\nInspirations, bits of code and source for being able to understand how Seq2Seq words or write my own Torch module come \nboth from [Ben Trevett](https://github.com/bentrevett/pytorch-seq2seq) and [Enrique Manjavacas](https://github.com/emanjavacas/pie). \n\n\n\n",
        "description_content_type": "text/markdown",
        "docs_url": null,
        "download_url": "",
        "downloads": {
            "last_day": -1,
            "last_month": -1,
            "last_week": -1
        },
        "home_page": "https://github.com/ponteineptique/boudams",
        "keywords": "",
        "license": "MIT",
        "maintainer": "",
        "maintainer_email": "",
        "name": "boudams",
        "package_url": "https://pypi.org/project/boudams/",
        "platform": "",
        "project_url": "https://pypi.org/project/boudams/",
        "project_urls": {
            "Homepage": "https://github.com/ponteineptique/boudams"
        },
        "release_url": "https://pypi.org/project/boudams/0.1.0/",
        "requires_dist": [
            "torch (>=1.0.1.post2)",
            "tqdm (>=4.31.1)",
            "unidecode (>=1.0.23)",
            "leven (==1.0.4)",
            "click (<=8.0,>=7.0)",
            "regex (==2019.5.25)",
            "mufidecode"
        ],
        "requires_python": ">=3.6.0",
        "summary": "A framework and toolkit for automatic segmentation",
        "version": "0.1.0"
    },
    "last_serial": 5963670,
    "releases": {
        "0.1.0": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "599f48648ae4fefab4d5af38d43fde98",
                    "sha256": "939726ac993eb910786a9599f7b714d4773bda12272f67b5ed0b278911780642"
                },
                "downloads": -1,
                "filename": "boudams-0.1.0-py2.py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "599f48648ae4fefab4d5af38d43fde98",
                "packagetype": "bdist_wheel",
                "python_version": "py2.py3",
                "requires_python": ">=3.6.0",
                "size": 3624,
                "upload_time": "2019-10-12T09:36:33",
                "url": "https://files.pythonhosted.org/packages/9e/6e/6fb5c75f8b14a6e1fb868d7c2790d4dfb587988cafc0152a195ba99e4990/boudams-0.1.0-py2.py3-none-any.whl"
            },
            {
                "comment_text": "",
                "digests": {
                    "md5": "bcbef31a04ff697bfac7c8c11632c6d7",
                    "sha256": "fac7046fb4ea673e0214c668555b9e98446d6cdaab61363d73207f2154fce311"
                },
                "downloads": -1,
                "filename": "boudams-0.1.0.tar.gz",
                "has_sig": false,
                "md5_digest": "bcbef31a04ff697bfac7c8c11632c6d7",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": ">=3.6.0",
                "size": 3978,
                "upload_time": "2019-10-12T09:36:35",
                "url": "https://files.pythonhosted.org/packages/a2/39/5e9a04e601004ca6942aefa1872852264187d67f0f28025a88c8a0574500/boudams-0.1.0.tar.gz"
            }
        ]
    },
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "md5": "599f48648ae4fefab4d5af38d43fde98",
                "sha256": "939726ac993eb910786a9599f7b714d4773bda12272f67b5ed0b278911780642"
            },
            "downloads": -1,
            "filename": "boudams-0.1.0-py2.py3-none-any.whl",
            "has_sig": false,
            "md5_digest": "599f48648ae4fefab4d5af38d43fde98",
            "packagetype": "bdist_wheel",
            "python_version": "py2.py3",
            "requires_python": ">=3.6.0",
            "size": 3624,
            "upload_time": "2019-10-12T09:36:33",
            "url": "https://files.pythonhosted.org/packages/9e/6e/6fb5c75f8b14a6e1fb868d7c2790d4dfb587988cafc0152a195ba99e4990/boudams-0.1.0-py2.py3-none-any.whl"
        },
        {
            "comment_text": "",
            "digests": {
                "md5": "bcbef31a04ff697bfac7c8c11632c6d7",
                "sha256": "fac7046fb4ea673e0214c668555b9e98446d6cdaab61363d73207f2154fce311"
            },
            "downloads": -1,
            "filename": "boudams-0.1.0.tar.gz",
            "has_sig": false,
            "md5_digest": "bcbef31a04ff697bfac7c8c11632c6d7",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": ">=3.6.0",
            "size": 3978,
            "upload_time": "2019-10-12T09:36:35",
            "url": "https://files.pythonhosted.org/packages/a2/39/5e9a04e601004ca6942aefa1872852264187d67f0f28025a88c8a0574500/boudams-0.1.0.tar.gz"
        }
    ]
}