{
    "info": {
        "author": "Ethan Rosenthal",
        "author_email": "ethanrosenthal@gmail.com",
        "bugtrack_url": null,
        "classifiers": [
            "Development Status :: 3 - Alpha",
            "Intended Audience :: Developers",
            "License :: OSI Approved :: MIT License",
            "Programming Language :: Python :: 3",
            "Topic :: Scientific/Engineering :: Information Analysis"
        ],
        "description": "# spacecutter\n\n`spacecutter` is a library for implementing ordinal regression models in PyTorch. The library consists of models and loss functions. It is recommended to use [skorch](http://skorch.readthedocs.io/) to wrap the models to make them compatible with scikit-learn.\n\n## Installation\n\nInstall locally via\n\n```bash\npip install -e .\n```\n\n## Usage\n\n### Models\n\nDefine any PyTorch model you want that generates a single, scalar prediction value. This will be our `predictor` model. This model can then be wrapped with `spacecutter.models.OrdinalLogisticModel` which will convert the output of the `predictor` from a single number to an array of ordinal class probabilities. The following example shows how to do this for a two layer neural network `predictor` for a problem with three ordinal classes.\n\n```python\nimport numpy as np\nimport torch\nfrom torch import nn\n\nfrom spacecutter.models import OrdinalLogisticModel\n\n\nX = np.array([[0.5, 0.1, -0.1],\n              [1.0, 0.2, 0.6],\n              [-2.0, 0.4, 0.8]],\n             dtype=np.float32)\n\ny = np.array([0, 1, 2]).reshape(-1, 1)\n\nnum_features = X.shape[1]\nnum_classes = len(np.unique(y))\n\npredictor = nn.Sequential(\n    nn.Linear(num_features, num_features),\n    nn.ReLU(),\n    nn.Linear(num_features, 1)\n)\n\nmodel = OrdinalLogisticModel(predictor, num_classes)\n\ny_pred = model(torch.as_tensor(X))\n\nprint(y_pred)\n\n# tensor([[0.2325, 0.2191, 0.5485],\n#         [0.2324, 0.2191, 0.5485],\n#         [0.2607, 0.2287, 0.5106]], grad_fn=<CatBackward>)\n\n```\n\n### Training\n\nIt is recommended to use [skorch](http://skorch.readthedocs.io/) to train `spacecutter` models. The following shows how to train the model from the previous section using cumulative link loss with `skorch`:\n\n```python\nfrom skorch import NeuralNet\n\nfrom spacecutter.callbacks import AscensionCallback\nfrom spacecutter.losses import CumulativeLinkLoss\n\nskorch_model = NeuralNet(\n    module=OrdinalLogisticModel,\n    module__predictor=predictor,\n    module__num_classes=num_classes,\n    criterion=CumulativeLinkLoss,\n    train_split=None,\n    callbacks=[\n        ('ascension', AscensionCallback()),\n    ],\n)\n\nskorch_model.fit(X, y)\n\n```\n\nNote that we must add the `AscensionCallback`. This ensures that the ordinal cutpoints stay in ascending order. While ideally this constraint would be factored directly into the model optimization, `spacecutter` currently hacks an SGD-compatible solution by utilizing a post-backwards-pass callback to clip the cutpoint values.\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/EthanRosenthal/spacecutter",
        "keywords": "machine learning statistics",
        "license": "",
        "maintainer": "",
        "maintainer_email": "",
        "name": "spacecutter",
        "package_url": "https://pypi.org/project/spacecutter/",
        "platform": "",
        "project_url": "https://pypi.org/project/spacecutter/",
        "project_urls": {
            "Homepage": "https://github.com/EthanRosenthal/spacecutter"
        },
        "release_url": "https://pypi.org/project/spacecutter/0.1.0/",
        "requires_dist": [
            "numpy (==1.15.4)",
            "torch (==0.4.1)",
            "flake8 (==3.6.0) ; extra == 'test'",
            "pytest (==4.0.1) ; extra == 'test'",
            "pytest-cov (==2.6.0) ; extra == 'test'",
            "skorch (==0.4.0) ; extra == 'test'"
        ],
        "requires_python": "",
        "summary": "Ordinal regression models in PyTorch",
        "version": "0.1.0"
    },
    "last_serial": 4566504,
    "releases": {
        "0.1.0": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "fde883705f4d890cdf96d78041f059bd",
                    "sha256": "3caa618fe3ff7b8f4dcb470a262cc52e5e10705472496f1eb8516a7cb009f4b1"
                },
                "downloads": -1,
                "filename": "spacecutter-0.1.0-py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "fde883705f4d890cdf96d78041f059bd",
                "packagetype": "bdist_wheel",
                "python_version": "py3",
                "requires_python": null,
                "size": 6654,
                "upload_time": "2018-12-06T05:03:50",
                "url": "https://files.pythonhosted.org/packages/1a/30/224d5e6733a35603171a56ca717487ebf42f8d62f5e84b9a9d772a2cfa80/spacecutter-0.1.0-py3-none-any.whl"
            },
            {
                "comment_text": "",
                "digests": {
                    "md5": "fe6f56caee5baad7f1328a76472266af",
                    "sha256": "b9e5d2e20951652d90253d2406abaa514ca8c50297a65498d5590987b4e8339f"
                },
                "downloads": -1,
                "filename": "spacecutter-0.1.0.tar.gz",
                "has_sig": false,
                "md5_digest": "fe6f56caee5baad7f1328a76472266af",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 5001,
                "upload_time": "2018-12-06T05:03:52",
                "url": "https://files.pythonhosted.org/packages/1b/8a/ebaf3ab290ae6207b556cf7bb969c73dd04c5e99a9403fd41ed5b401d1f8/spacecutter-0.1.0.tar.gz"
            }
        ]
    },
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "md5": "fde883705f4d890cdf96d78041f059bd",
                "sha256": "3caa618fe3ff7b8f4dcb470a262cc52e5e10705472496f1eb8516a7cb009f4b1"
            },
            "downloads": -1,
            "filename": "spacecutter-0.1.0-py3-none-any.whl",
            "has_sig": false,
            "md5_digest": "fde883705f4d890cdf96d78041f059bd",
            "packagetype": "bdist_wheel",
            "python_version": "py3",
            "requires_python": null,
            "size": 6654,
            "upload_time": "2018-12-06T05:03:50",
            "url": "https://files.pythonhosted.org/packages/1a/30/224d5e6733a35603171a56ca717487ebf42f8d62f5e84b9a9d772a2cfa80/spacecutter-0.1.0-py3-none-any.whl"
        },
        {
            "comment_text": "",
            "digests": {
                "md5": "fe6f56caee5baad7f1328a76472266af",
                "sha256": "b9e5d2e20951652d90253d2406abaa514ca8c50297a65498d5590987b4e8339f"
            },
            "downloads": -1,
            "filename": "spacecutter-0.1.0.tar.gz",
            "has_sig": false,
            "md5_digest": "fe6f56caee5baad7f1328a76472266af",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": null,
            "size": 5001,
            "upload_time": "2018-12-06T05:03:52",
            "url": "https://files.pythonhosted.org/packages/1b/8a/ebaf3ab290ae6207b556cf7bb969c73dd04c5e99a9403fd41ed5b401d1f8/spacecutter-0.1.0.tar.gz"
        }
    ]
}