{
    "info": {
        "author": "Valentin Lievin",
        "author_email": "valentin.lievin@gmail.com",
        "bugtrack_url": null,
        "classifiers": [
            "License :: OSI Approved :: MIT License",
            "Operating System :: OS Independent",
            "Programming Language :: Python :: 3"
        ],
        "description": "# Booster\n\nA lightweight library to ease the training and the debugging of deep neural networks with PyTorch. Data structures and paradigms.\n\n## Data Structures\n\n### Diagnostic\n\nA two level dictionary structure to store the model diagnostics. Compatible with Tensorboard datastructure.\n\nExample:\n\n```python\nfrom booster import Diagnostic\n\ndata = {\n'loss' : {'nll' : [45., 58.], 'kl': [22., 18.]},\n'info' : {'batch_size' : 16, 'runtime' : 0.01}\n}\n\ndiagnostic = Diagnostic(data)\n```\n\n### Aggregator\n\nA module to compute the running average of the diagnostics.\n\n```python\nfrom booster import Aggregator, Diagnostic\n\naggregator = Aggregator()\n...\naggregator.initialize()\nfor x in data_loader:\n  data = optimization_step(model, data)\n  aggregator.update(data)\n\nsummmary = aggregator.data # summary is an instance of Diagnostic\nsummmary = summary.to('cpu')\n```\n\nThe output is a Diagnostic object and can easily be logged to Tensorboard.\n\n```python\n# log to tensorboard\nwriter = SummaryWriter(log_dir=\"...\")\nsummary.log(writer, global_step)\n\n```\n\n## Evaluator\n\nThe Evaluator computes a forward pass through the model, the loss and additional Diagnostics.\n\n```python\nfrom booster.evaluation import Classification\nmodel = Classifier()\nevaluator = Classification(categories=10)\n\n# evaluate model\ndata = next(iter(loader))\nloss, diagnostics, output = evaluator(model, data)\n\n```\n\n## Pipeline: model + evaluator\n \nThe pipeline fuses the model forward pass with the evaluator and can be wrapped into a custom Dataparallel class that handles the diagnostics.\n\n```python\nfrom booster import Pipeline, DataParallelPipeline\n\n# fuse model + evaluator\npipeline = Pipeline(model, evaluator)\n\n# wrap as DataParallel\nparallel_pipeline = DataParallelPipeline(pipeline, device_ids=device_ids)\n\n# evaluate model on multiple devices and gather loss and diagnostics\ndata = next(iter(loader))\nloss, diagnostics, output = parallel_pipeline(data) \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/vlievin/booster-pytorch",
        "keywords": "",
        "license": "",
        "maintainer": "",
        "maintainer_email": "",
        "name": "booster-pytorch",
        "package_url": "https://pypi.org/project/booster-pytorch/",
        "platform": "",
        "project_url": "https://pypi.org/project/booster-pytorch/",
        "project_urls": {
            "Homepage": "https://github.com/vlievin/booster-pytorch"
        },
        "release_url": "https://pypi.org/project/booster-pytorch/0.0.2/",
        "requires_dist": null,
        "requires_python": "",
        "summary": "A lightweight library to ease the training and the debugging of deep neural networks with PyTorch. Data structures and paradigms.",
        "version": "0.0.2",
        "yanked": false,
        "yanked_reason": null
    },
    "last_serial": 7534375,
    "releases": {
        "0.0.1": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "a733190dc8e3f5c13fc80fbf776e1a83",
                    "sha256": "8aefb907553abe188f68bb6e98f9302702c3ef475db6570d218b67a297c0a8d1"
                },
                "downloads": -1,
                "filename": "booster-pytorch-0.0.1.tar.gz",
                "has_sig": false,
                "md5_digest": "a733190dc8e3f5c13fc80fbf776e1a83",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 9284,
                "upload_time": "2019-10-30T10:03:21",
                "upload_time_iso_8601": "2019-10-30T10:03:21.930778Z",
                "url": "https://files.pythonhosted.org/packages/b1/cb/3c6fa9efe2ed9c997abd2507062193f42be9cb1d6ae89de9ec5e3bf100a3/booster-pytorch-0.0.1.tar.gz",
                "yanked": false,
                "yanked_reason": null
            }
        ],
        "0.0.2": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "8c42604367c2e01e9c43fa3e3029adad",
                    "sha256": "b773b4afb42db509a8417157ee85c2af7f240ab09bec5e958a7ea69b601c1ff9"
                },
                "downloads": -1,
                "filename": "booster-pytorch-0.0.2.tar.gz",
                "has_sig": false,
                "md5_digest": "8c42604367c2e01e9c43fa3e3029adad",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 14205,
                "upload_time": "2020-06-22T14:02:58",
                "upload_time_iso_8601": "2020-06-22T14:02:58.208055Z",
                "url": "https://files.pythonhosted.org/packages/0d/96/4725f89b367c3649dac69af0de4ee2d230cd695d85038cae02d9d2cc039a/booster-pytorch-0.0.2.tar.gz",
                "yanked": false,
                "yanked_reason": null
            }
        ]
    },
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "md5": "8c42604367c2e01e9c43fa3e3029adad",
                "sha256": "b773b4afb42db509a8417157ee85c2af7f240ab09bec5e958a7ea69b601c1ff9"
            },
            "downloads": -1,
            "filename": "booster-pytorch-0.0.2.tar.gz",
            "has_sig": false,
            "md5_digest": "8c42604367c2e01e9c43fa3e3029adad",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": null,
            "size": 14205,
            "upload_time": "2020-06-22T14:02:58",
            "upload_time_iso_8601": "2020-06-22T14:02:58.208055Z",
            "url": "https://files.pythonhosted.org/packages/0d/96/4725f89b367c3649dac69af0de4ee2d230cd695d85038cae02d9d2cc039a/booster-pytorch-0.0.2.tar.gz",
            "yanked": false,
            "yanked_reason": null
        }
    ],
    "vulnerabilities": []
}