{
    "info": {
        "author": "Dylan Albrecht",
        "author_email": "djalbrecht@email.wm.edu",
        "bugtrack_url": null,
        "classifiers": [
            "Intended Audience :: Developers",
            "Intended Audience :: Science/Research",
            "License :: OSI Approved",
            "Operating System :: MacOS",
            "Operating System :: Microsoft :: Windows",
            "Operating System :: POSIX",
            "Operating System :: Unix",
            "Programming Language :: Python",
            "Programming Language :: Python :: 3.7",
            "Topic :: Scientific/Engineering",
            "Topic :: Software Development"
        ],
        "description": "![fast](images/leapy.gif)\n\n**Just give me the dataz!**\n\nWelcome!  Leapy is a library for real-time, sub-millisecond inference;\nit provides customizable machine learning pipeline export for fast\nmodel serving. \nThese pipelines are targeted for using Dask's scalable machine learning,\nwhich follows the Scikit-Learn API. However, you can use this framework\ndirectly with Scikit-Learn as well.\n\n```python\npipe = Pipeline([\n        ('fp', FeaturePipeline([('ohe',\n                                 OneHotEncoder(sparse=False),\n                                 [0, 1])])),\n        ('clf', LogisticRegression())\n])\n\npipe.fit(X, y)\n\npipeline_runtime = pipe.to_runtime()               # \u26a1\u26a1\u26a1 \ninit('./model_repo', pipeline_runtime, df.head())  # Ready to deploy\n```\n\nAnd serve this super fast pipeline:\n```\n$ leap serve --repo ./model_repo\n$ curl localhost:8080:/health\n{\n  \"status\": \"healthy\"\n}\n```\n(See below for benchmarks and a more detailed usage example.)\n\n### Benefits\n\nDask is a Python distributed computing environment in Python with a\nburgeoning machine learning component, compatible with Scikit-Learn's API.\nUsing Leapy's framework, we can serve these pipelines in real-time!\n\nThis means:\n\n* No JVM: No reliance on JVM from using Spark.\n* Python: All Python development and custom transformers -- no Scala & Java\n          needed!\n* Scale: Scikit-Learn logic and pipelines scaled to clusters.\n* Fast: You're in control of how fast your transformers are.\n* Simple: Easily build and deploy models with Docker.\n* Reliable: Encourages a test-driven approach.\n<!--* MLflow: Serve runtime models (as Scikit-Learn models) through `mlflow`.-->\n\n### Examples\n\n* [Simple](examples/simple)\n    -- Super simple example of creating, testing, and using custom\n    transformers\n* [XGBoost](examples/)\n    -- Advanced example of using XGBoost with Dask, saving, and serving the\n    model.\n\n### Benchmarks\n\nA simple example of what we're going for -- computing a one-hot-encoding,\nwith ~200K labels, of a single data point (dask array `x_da` and numpy array\n`x = x_da.compute()`):\n\n![sample benchmark](images/sample_benchmark.png)\n\nWhere `ohe_dml` (from `dask_ml`) and `ohe` (from `leapy`) are essentially the\nsame; `ohe_sk` is from `scikit-learn` and `ohe_runtime` is from\n`ohe.to_runtime()`. And, running `compute()` on Dask transforms above bumps\nthe time up to about 1 second.\n\nWith the time we save using `ohe_runtime`, we can squeeze in many more\ntransformations and an estimator to still come in under 1ms.\n\n### Example Usage\n\nStart with a dataset in dask arrays, `X`, `y`, and dataframe `ddf`:\npipeline:\n```python\nimport numpy as np\nimport pandas as pd\nimport dask.array as da\nimport dask.dataframe as dd\n\nX_np = np.array([[1, 'a'], [2, 'b']], dtype=np.object)\ndf_pd = pd.DataFrame(X_np, columns=['test_int', 'test_str'])\ny_np = np.array([0, 1])\n\nX = da.from_array(X_np, chunks=X_np.shape)\ny = da.from_array(y_np, chunks=y_np.shape)\nddf = dd.from_pandas(df_pd, npartitions=1)\n```\n\nCreate our pipeline:\n\n```python\nfrom sklearn.pipeline import Pipeline\nfrom dask_ml.linear_model import LogisticRegression\nfrom leapy.dask.transformers import OneHotEncoder\nfrom leapy.dask.pipeline import FeaturePipeline\nfrom leapy.serve import init\n\npipe = Pipeline([\n        ('fp', FeaturePipeline([\n            # One-Hot-Encode 'test_str' feature, drop 'test_int'\n            ('ohe', OneHotEncoder(sparse=False), [1])])),\n        ('clf', LogisticRegression())\n])\n\npipe.fit(X, y)\n```\n\nThen we export to a runtime pipeline and get ready for model serving:\n\n```python\npipe_runtime = pipe.to_runtime()\ninit('./model_repo', pipe_runtime, ddf.head())\n```\n\nFinally we serve the model:\n```\n$ leap serve --repo ./model_repo\n$ curl localhost:8080/predict \\\n    -X POST \\\n    -H \"Content-Type: application/json\" \\\n    --data '{\"test_int\": 1, \"test_str\": \"b\"}'\n{\n  \"prediction\": 1.0\n}\n```\n\n\nFor more on model serving see [leapy/serve/README.md](leapy/serve/README.md).\n\n## Acknowledgments\n\nLeapy is inspired by [MLeap](https://github.com/combust/mleap).\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": "http://github.com/nonabelian/leapy",
        "keywords": "",
        "license": "new BSD",
        "maintainer": "",
        "maintainer_email": "",
        "name": "leapy",
        "package_url": "https://pypi.org/project/leapy/",
        "platform": "",
        "project_url": "https://pypi.org/project/leapy/",
        "project_urls": {
            "Homepage": "http://github.com/nonabelian/leapy"
        },
        "release_url": "https://pypi.org/project/leapy/0.1.0/",
        "requires_dist": [
            "pandas (>=0.24.2)",
            "dask (>=1.1.4)",
            "dask-ml (>=0.12.0)",
            "numpy (>=1.16.2)",
            "scikit-learn (>=0.20.3)",
            "marshmallow (>=2.19.2)",
            "docker (==3.7.2)"
        ],
        "requires_python": "",
        "summary": "Real-time inference pipelines",
        "version": "0.1.0"
    },
    "last_serial": 5139817,
    "releases": {
        "0.0.2": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "7054f9d2c6196fc2a5d70a255b1afe94",
                    "sha256": "bbcc02ef254e8a5efc7339bbeb483c44e506b8fe6d1dec1353405c5abf9c4108"
                },
                "downloads": -1,
                "filename": "leapy-0.0.2-py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "7054f9d2c6196fc2a5d70a255b1afe94",
                "packagetype": "bdist_wheel",
                "python_version": "py3",
                "requires_python": null,
                "size": 19421,
                "upload_time": "2019-04-10T01:05:32",
                "url": "https://files.pythonhosted.org/packages/9e/d8/91a8c5048a5391e00cfa07952a04f62dfd91ab07189947b64a4814b61813/leapy-0.0.2-py3-none-any.whl"
            }
        ],
        "0.0.3": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "74077f375afd35cdd335f9d9c513848a",
                    "sha256": "a4a64cec1458298dfd49748ed85e394ae524f808aea79c37da746511d4bcb373"
                },
                "downloads": -1,
                "filename": "leapy-0.0.3-py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "74077f375afd35cdd335f9d9c513848a",
                "packagetype": "bdist_wheel",
                "python_version": "py3",
                "requires_python": null,
                "size": 20207,
                "upload_time": "2019-04-14T01:55:49",
                "url": "https://files.pythonhosted.org/packages/52/88/198f3fe2a29c83d7276920fde78aba6aa5471a324e12037bccda8a328fb9/leapy-0.0.3-py3-none-any.whl"
            },
            {
                "comment_text": "",
                "digests": {
                    "md5": "64288db85895bf62b1d1fe611e833617",
                    "sha256": "0f023f28ba4e546b2e9066a1c2d3ba68fe119b3f2b2854ff41fc44161bc79ac9"
                },
                "downloads": -1,
                "filename": "leapy-0.0.3.tar.gz",
                "has_sig": false,
                "md5_digest": "64288db85895bf62b1d1fe611e833617",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 13133,
                "upload_time": "2019-04-14T01:55:51",
                "url": "https://files.pythonhosted.org/packages/0f/8c/86407433706e84e22eaeb71ec9a60787f75b7ddc7da8f9d023c626126727/leapy-0.0.3.tar.gz"
            }
        ],
        "0.1.0": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "3cf01cae541ffe754699127296fecb3a",
                    "sha256": "c92a7d0c6ae90fa341b2a358e014bdc0d4850da8b4214757e63f1675c8ce4ae5"
                },
                "downloads": -1,
                "filename": "leapy-0.1.0-py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "3cf01cae541ffe754699127296fecb3a",
                "packagetype": "bdist_wheel",
                "python_version": "py3",
                "requires_python": null,
                "size": 22964,
                "upload_time": "2019-04-14T04:07:46",
                "url": "https://files.pythonhosted.org/packages/f5/16/cccd9d3210e1c1f7123ad11f59a2ebab800c4c484c9b657ffd80a3c0ac6b/leapy-0.1.0-py3-none-any.whl"
            },
            {
                "comment_text": "",
                "digests": {
                    "md5": "797476fad7858de66a6766560795c6af",
                    "sha256": "d7cc0102c8902b3c6364c66356d1aa3fee1387033ced01143b6e90901ae92a94"
                },
                "downloads": -1,
                "filename": "leapy-0.1.0.tar.gz",
                "has_sig": false,
                "md5_digest": "797476fad7858de66a6766560795c6af",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 15836,
                "upload_time": "2019-04-14T04:07:48",
                "url": "https://files.pythonhosted.org/packages/dd/23/28e3c4af3b633f5b4f4241818babf8d3f6820b88e4e7c4ddd6640898912b/leapy-0.1.0.tar.gz"
            }
        ]
    },
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "md5": "3cf01cae541ffe754699127296fecb3a",
                "sha256": "c92a7d0c6ae90fa341b2a358e014bdc0d4850da8b4214757e63f1675c8ce4ae5"
            },
            "downloads": -1,
            "filename": "leapy-0.1.0-py3-none-any.whl",
            "has_sig": false,
            "md5_digest": "3cf01cae541ffe754699127296fecb3a",
            "packagetype": "bdist_wheel",
            "python_version": "py3",
            "requires_python": null,
            "size": 22964,
            "upload_time": "2019-04-14T04:07:46",
            "url": "https://files.pythonhosted.org/packages/f5/16/cccd9d3210e1c1f7123ad11f59a2ebab800c4c484c9b657ffd80a3c0ac6b/leapy-0.1.0-py3-none-any.whl"
        },
        {
            "comment_text": "",
            "digests": {
                "md5": "797476fad7858de66a6766560795c6af",
                "sha256": "d7cc0102c8902b3c6364c66356d1aa3fee1387033ced01143b6e90901ae92a94"
            },
            "downloads": -1,
            "filename": "leapy-0.1.0.tar.gz",
            "has_sig": false,
            "md5_digest": "797476fad7858de66a6766560795c6af",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": null,
            "size": 15836,
            "upload_time": "2019-04-14T04:07:48",
            "url": "https://files.pythonhosted.org/packages/dd/23/28e3c4af3b633f5b4f4241818babf8d3f6820b88e4e7c4ddd6640898912b/leapy-0.1.0.tar.gz"
        }
    ]
}