{
    "info": {
        "author": "shobrook",
        "author_email": "shobrookj@gmail.com",
        "bugtrack_url": null,
        "classifiers": [],
        "description": "# sequitur\n\n`sequitur` is a Recurrent Autoencoder (RAE) for sequence data that works out-of-the-box. It's easy to configure and only takes one line of code to use.\n\n```python\nfrom sequitur import QuickEncode\n\nsequences = [[1,2,3,4], [5,6,7,8], [9,10,11,12]]\nencoder, decoder, embeddings, f_loss  = QuickEncode(sequences, embedding_dim=2)\n\nencoder([13,14,15,16]) # => [0.19, 0.84]\n```\n\n<!--An autoencoder will learn how to-->`sequitur` will learn how to represent sequences of any length as lower-dimensional, fixed-size vectors. This can be useful for finding patterns among sequences, clustering, converting sequences into inputs for a machine learning algorithm, and dimensionality reduction.\n\n## Installation\n\n> Requires Python 3.X and PyTorch 1.2.X\n\nYou can download a compiled binary [here](https://github.com/shobrook/sequitur/) or install `sequitur` with pip:\n\n`$ pip install sequitur`\n\n## API\n\n#### `sequitur.QuickEncode(sequences, embedding_dim, logging=False, lr=1e-3, epochs=100)`\n\nLets you train an autoencoder with just one line of code. This wraps a PyTorch implementation of an Encoder-Decoder architecture with an LSTM, making this optimal for sequences with long-term dependencies (e.g. time series data).\n\n**Parameters**\n\n- `sequences`: A list (or tensor) of shape `[num_seqs, seq_len, num_features]` representing your training set of sequences.\n  - Each sequence should have the same length, `seq_len`, and contain a sequence of vectors of size `num_features`.\n  - If `num_features=1`, then you can input a list of shape `[num_seqs, seq_len]` instead.\n- `embedding_dim`: Size of the vector encodings you want to create.\n- `logging`: Boolean for whether you want logging statements to be printed during training.\n- `lr`: Learning rate for the autoencoder.\n- `epochs`: Number of epochs to train for.\n\n**Returns**\n\n- `encoder`: The trained encoder as a PyTorch module.\n  - Takes as input a tensor of shape `[seq_len, num_features]` representing a sequence where each element is a vector of size `num_features`.\n- `decoder`: The trained decoder as a PyTorch module.\n  - Takes as input a tensor of shape `[embedding_dim]` representing an encoded sequence.\n- `embeddings`: A tensor of shape `[num_seqs, embedding_dim]` which holds the learned vector encodings of each sequence in the training set.\n- `f_loss`: The final mean squared error of the autoencoder on the training set.\n\n#### `sequitur.autoencoders.RAE(hyperparams)`\n\nTo-Do.\n\n#### `sequitur.autoencoders.SAE(hyperparams)`\n\nTo-Do.\n\n### `sequitur.autoencoders.VAE(hyperparams)`\n\nTo-Do.\n\n## Contributing\n\n`QuickEncode` is useful for rapid prototyping but doesn't give you much control over the model and training process. For that, you can import the RAE implementation itself from `sequitur.autoencoders`.\n\n`sequitur` not only implements an RAE but also a Stacked Autoencoder (SAE) and a WIP Variational Autoencoder (VAE). If you've implemented a sequence autoencoder, or know of an implementation, please feel free to add it to the codebase and open a pull request. With enough autoencoders, I can turn `sequitur` into a small PyTorch extension library.\n\n<!--Provide proof that it's generally effective-->\n\n<!-- https://github.com/szagoruyko/pytorchviz\nhttps://github.com/RobRomijnders/AE_ts\nhttps://github.com/erickrf/autoencoder\nhttps://miro.medium.com/max/1400/1*sWc8g2yiQrOzntbVeGzbEQ.png\nhttps://arxiv.org/pdf/1502.04681.pdf -->",
        "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/shobrook/sequitur",
        "keywords": "sequitur,autoencoder,lstm,sequence,sequence-data,seq2seq",
        "license": "MIT",
        "maintainer": "",
        "maintainer_email": "",
        "name": "sequitur",
        "package_url": "https://pypi.org/project/sequitur/",
        "platform": "",
        "project_url": "https://pypi.org/project/sequitur/",
        "project_urls": {
            "Homepage": "https://github.com/shobrook/sequitur"
        },
        "release_url": "https://pypi.org/project/sequitur/1.0.2/",
        "requires_dist": null,
        "requires_python": ">=3",
        "summary": "Recurrent Autoencoder for sequence data that works out-of-the-box",
        "version": "1.0.2"
    },
    "last_serial": 5901331,
    "releases": {
        "1.0.0": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "10397eb4cd53f7d1a297bae33f4dd3cd",
                    "sha256": "e9a1da1853efa731c85e8e341d312b0c366fd569195d7c2c9bec188cd76ad3e9"
                },
                "downloads": -1,
                "filename": "sequitur-1.0.0.tar.gz",
                "has_sig": false,
                "md5_digest": "10397eb4cd53f7d1a297bae33f4dd3cd",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": ">=3",
                "size": 3436,
                "upload_time": "2019-09-29T03:01:44",
                "url": "https://files.pythonhosted.org/packages/89/a1/d2c01bab2321c43d378fcf07930badf119d0bf7569355f6f3548639a589d/sequitur-1.0.0.tar.gz"
            }
        ],
        "1.0.1": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "4917126caa71ce9621577bd543a9964d",
                    "sha256": "36c5103df8bab6561d21c64308c0bbae4e1911ba45e636a3e3f51c88adc84eab"
                },
                "downloads": -1,
                "filename": "sequitur-1.0.1.tar.gz",
                "has_sig": false,
                "md5_digest": "4917126caa71ce9621577bd543a9964d",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": ">=3",
                "size": 3438,
                "upload_time": "2019-09-29T03:04:32",
                "url": "https://files.pythonhosted.org/packages/26/c5/c4249dc500d8ed37a0d6c1ab87c81be83b2f9be71af3b6fb35a58bdc0a81/sequitur-1.0.1.tar.gz"
            }
        ],
        "1.0.2": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "16d899abfd714263857d6c47969b8843",
                    "sha256": "dab6b8c551ce4b12e8fdd912e54a10a1de406fd49fd791904179293fb71a9f83"
                },
                "downloads": -1,
                "filename": "sequitur-1.0.2.tar.gz",
                "has_sig": false,
                "md5_digest": "16d899abfd714263857d6c47969b8843",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": ">=3",
                "size": 3852,
                "upload_time": "2019-09-29T03:08:06",
                "url": "https://files.pythonhosted.org/packages/68/de/6f730755810d3e115b196f345d14b59e014a672602a522975032c06969bf/sequitur-1.0.2.tar.gz"
            }
        ]
    },
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "md5": "16d899abfd714263857d6c47969b8843",
                "sha256": "dab6b8c551ce4b12e8fdd912e54a10a1de406fd49fd791904179293fb71a9f83"
            },
            "downloads": -1,
            "filename": "sequitur-1.0.2.tar.gz",
            "has_sig": false,
            "md5_digest": "16d899abfd714263857d6c47969b8843",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": ">=3",
            "size": 3852,
            "upload_time": "2019-09-29T03:08:06",
            "url": "https://files.pythonhosted.org/packages/68/de/6f730755810d3e115b196f345d14b59e014a672602a522975032c06969bf/sequitur-1.0.2.tar.gz"
        }
    ]
}