{
    "info": {
        "author": "Michael Fuerst",
        "author_email": "mail@michaelfuerst.de",
        "bugtrack_url": null,
        "classifiers": [
            "Development Status :: 3 - Alpha",
            "Intended Audience :: Developers",
            "Programming Language :: Python"
        ],
        "description": "# starttf - Simplified Deeplearning for Tensorflow [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)\n\nThis repo aims to contain everything required to quickly develop a deep neural network with tensorflow.\nThe idea is that if you use write a compatible SimpleSequence for data loading and networks based on the StartTFModel, you will automatically obey best practices and have super fast training speeds.\n\n## Install\n\nProperly install `tensorflow` or `tensorflow-gpu` please follow the [official instructions](https://www.tensorflow.org/install/) carefully.\n\nThen, simply pip install from the github repo.\n\n```bash\npip install starttf\n```\n\n## Datasets\n\nExtensions SimpleSequences from [opendatalake.simple_sequence.SimpleSequence](https://github.com/penguinmenac3/opendatalake/blob/master/opendatalake/simple_sequence.py) are supported.\nThey work like keras.Sequence however with an augmentation and a preprocessing function.\n\nFor details checkout the [readme of opendatalake](https://github.com/penguinmenac3/opendatalake/blob/master/README.md).\n\n## Models\n\nEvery model returns a dictionary containing output tensors and a dictionary containing debug tensors\n\n1. [Model Base Classes](starttf/models/models.py)\n2. [Common Encoders](starttf/models/encoders.py)\n3. [Untrained Backbones](starttf/models/backbones)\n\n## Simple to use tensorflow\n\n### Simple Training (No Boilerplate)\n\nThere are pre-implemented models which can be glued together and trained with just a few lines.\nHowever, before training you will have to create tf-records as shown in the section *Simple TF Record Creation*.\nThis is actually a full main file.\n\n```python\n# Import helpers\nfrom starttf.estimators.tf_estimator import easy_train_and_evaluate\nfrom starttf.utils.hyperparams import load_params\n\n# Import a/your model (here one for mnist)\nfrom mymodel import MyStartTFModel\n\n# Import your loss (here an example)\nfrom myloss import create_loss\n\n# Load params (here for mnist)\nhyperparams = load_params(\"hyperparams/experiment1.json\")\n\n# Train model\neasy_train_and_evaluate(hyperparams, MyStartTFModel, create_loss, continue_training=False)\n```\n\n### Quick Model Definition\n\nSimply implement a create_model function.\nThis model is only a feed forward model.\n\nThe model function returns a dictionary containing all layers that should be accessible from outside and a dictionary containing debug values that should be availible for loss or plotting in tensorboard.\n\n```python\nimport tensorflow as tf\n\nfrom starttf.models.model import StartTFModel\nfrom starttf.models.encoders import Encoder\n\nConv2D = tf.keras.layers.Conv2D\n\n\nclass ExampleModel(StartTFModel):\n    def __init__(self, hyperparams):\n        super(ExampleModel, self).__init__(hyperparams)\n        num_classes = hyperparams.problem.number_of_categories\n\n        # Create the vgg encoder\n        self.encoder = Encoder(hyperparams)\n\n        #Use the generated model \n        self.conv6 = Conv2D(filters=1024, kernel_size=(1, 1), padding=\"same\", activation=\"relu\")\n        self.conv7 = Conv2D(filters=1024, kernel_size=(1, 1), padding=\"same\", activation=\"relu\")\n        self.conv8 = Conv2D(filters=num_classes, kernel_size=(1, 1), padding=\"same\", activation=None, name=\"probs\")\n\n    def call(self, input_tensor, training=False):\n        \"\"\"\n        Run the model.\n        \"\"\"\n        encoder, debug = self.encoder(input_tensor, training)\n        result = self.conv6(encoder[\"features\"])\n        result = self.conv7(result)\n        logits = self.conv8(result)\n        probs = tf.nn.softmax(logits)\n        return {\"logits\": logits, \"probs\": probs}, debug\n```\n\n### Quick Loss Definition\n\n```python\ndef create_loss(model, labels, mode, hyper_params):\n    metrics = {}\n    losses = {}\n\n    # Add loss\n    labels = tf.reshape(labels[\"probs\"], [-1, hyper_params.problem.number_of_categories])\n    ce = tf.nn.softmax_cross_entropy_with_logits_v2(logits=model[\"logits\"], labels=labels)\n    loss_op = tf.reduce_mean(ce)\n\n    # Add losses to dict. \"loss\" is the primary loss that is optimized.\n    losses[\"loss\"] = loss_op\n    metrics['accuracy'] = tf.metrics.accuracy(labels=labels,\n                                              predictions=model[\"probs\"],\n                                              name='acc_op')\n\n    return losses, metrics\n```\n\n### Simple TF Record Creation\n\nFast training speed can be achieved by using tf records.\nHowever, usually tf records are a hastle to use the write_data method makes it simple.\n\n```python\nfrom starttf.utils.hyperparams import load_params\nfrom starttf.data.autorecords import write_data\n\nfrom my_data import MySimpleSequence\n\n# Load the hyper parameters.\nhyperparams = load_params(\"hyperparams/experiment1.json\")\n\n# Get a generator and its parameters\ntraining_data = MySimpleSequence(hyperparams)\nvalidation_data = MySimpleSequence(hyperparams)\n\n# Write the data\nwrite_data(hyperparams, PHASE_TRAIN, training_data, 4)\nwrite_data(hyperparams, PHASE_VALIDATION, validation_data, 2)\n```\n\n### Tensorboard Integration\n\nTensorboard integration is simple.\n\nEvery loss in the losses dict is automatically added to tensorboard.\nIf you also want debug images, you can add a tf.summary.image() in your create_loss method.\n\n### TF Estimator + Cluster Support\n\nIf you use the easy_train_and_evaluate method, a correctly configured TF Estimator is created.\nThe estimator is then trained in a way that supports cluster training if you have a cluster.\n\n\n",
        "description_content_type": "",
        "docs_url": null,
        "download_url": "https://github.com/penguinmenac3/starttf/tarball/1.0.3",
        "downloads": {
            "last_day": -1,
            "last_month": -1,
            "last_week": -1
        },
        "home_page": "https://github.com/penguinmenac3/starttf",
        "keywords": "",
        "license": "BSD",
        "maintainer": "",
        "maintainer_email": "",
        "name": "starttf",
        "package_url": "https://pypi.org/project/starttf/",
        "platform": "",
        "project_url": "https://pypi.org/project/starttf/",
        "project_urls": {
            "Download": "https://github.com/penguinmenac3/starttf/tarball/1.0.3",
            "Homepage": "https://github.com/penguinmenac3/starttf"
        },
        "release_url": "https://pypi.org/project/starttf/1.0.3/",
        "requires_dist": [
            "numpy",
            "scikit-image",
            "scikit-learn",
            "scipy",
            "matplotlib",
            "h5py",
            "pandas",
            "opencv-python",
            "tensorflow-hub",
            "hyperparams",
            "opendatalake",
            "GPUtil",
            "keras"
        ],
        "requires_python": "",
        "summary": "A tensorflow batteries included kit to write tensorflow networks from scratch or use existing ones.",
        "version": "1.0.3"
    },
    "last_serial": 4481866,
    "releases": {
        "0.0.2": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "b50683d6fb7e84cfe0bd208cdf85f20f",
                    "sha256": "8fbf2fd7b48f72b54c7a0f9f01d127bc5d711528c25f4bed996f63c5b9b690d0"
                },
                "downloads": -1,
                "filename": "starttf-0.0.2.tar.gz",
                "has_sig": false,
                "md5_digest": "b50683d6fb7e84cfe0bd208cdf85f20f",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 40353,
                "upload_time": "2018-05-12T17:55:35",
                "url": "https://files.pythonhosted.org/packages/96/be/5d00f2aae8fcddf53e4bcd34c09a38469dec3d49e11eabaf0e122f03fd2a/starttf-0.0.2.tar.gz"
            }
        ],
        "0.0.3": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "78ff1f5b309d49b24068381cccff52e6",
                    "sha256": "8c01517cf54a66a832e8585fbca09f803e5759f4c4b4d1770c9a0de8c6a0ab8e"
                },
                "downloads": -1,
                "filename": "starttf-0.0.3.tar.gz",
                "has_sig": false,
                "md5_digest": "78ff1f5b309d49b24068381cccff52e6",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 40447,
                "upload_time": "2018-05-23T13:06:34",
                "url": "https://files.pythonhosted.org/packages/46/90/35582fde84554348341fa91948dbadb6f02a3b4443a68c83d84b048d383e/starttf-0.0.3.tar.gz"
            }
        ],
        "1.0.0": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "b3b04fb70c5bb5cc1a7bb4ef18711520",
                    "sha256": "de132f40e0fa1e7516f9c0c5155a1cae6d0ac5bc7372b654fce0687b0bfddeb0"
                },
                "downloads": -1,
                "filename": "starttf-1.0.0-py2.py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "b3b04fb70c5bb5cc1a7bb4ef18711520",
                "packagetype": "bdist_wheel",
                "python_version": "py2.py3",
                "requires_python": null,
                "size": 64355,
                "upload_time": "2018-10-26T18:00:32",
                "url": "https://files.pythonhosted.org/packages/11/7a/546c19740a1797b0a72c9098d1ee1464decf32ad28eaa0c842bf625b67b9/starttf-1.0.0-py2.py3-none-any.whl"
            }
        ],
        "1.0.1": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "e71344826313c5f1f40782c22077b5b1",
                    "sha256": "2e94467fbb2883bf52577f05a5fefd861700c6cc1732cdf493d79c9c372090c7"
                },
                "downloads": -1,
                "filename": "starttf-1.0.1-py2.py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "e71344826313c5f1f40782c22077b5b1",
                "packagetype": "bdist_wheel",
                "python_version": "py2.py3",
                "requires_python": null,
                "size": 64487,
                "upload_time": "2018-10-30T10:50:34",
                "url": "https://files.pythonhosted.org/packages/12/20/f75c70dc1706e24303db3265fb03eb175478f16c86b966b2771964e9d124/starttf-1.0.1-py2.py3-none-any.whl"
            }
        ],
        "1.0.3": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "1c58017c49e0c6417bf27b3f8f86f51c",
                    "sha256": "c5b8bc66b0ea9d2e84b2826bfe9eb675aeeeeb8c94a1d4ce98ff2da2fe25e0e2"
                },
                "downloads": -1,
                "filename": "starttf-1.0.3-py2.py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "1c58017c49e0c6417bf27b3f8f86f51c",
                "packagetype": "bdist_wheel",
                "python_version": "py2.py3",
                "requires_python": null,
                "size": 64695,
                "upload_time": "2018-11-13T14:05:25",
                "url": "https://files.pythonhosted.org/packages/62/67/d2b73956ef790b9b42ab88776de99e8e046bceec6e1294fec94ea3c57f61/starttf-1.0.3-py2.py3-none-any.whl"
            }
        ]
    },
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "md5": "1c58017c49e0c6417bf27b3f8f86f51c",
                "sha256": "c5b8bc66b0ea9d2e84b2826bfe9eb675aeeeeb8c94a1d4ce98ff2da2fe25e0e2"
            },
            "downloads": -1,
            "filename": "starttf-1.0.3-py2.py3-none-any.whl",
            "has_sig": false,
            "md5_digest": "1c58017c49e0c6417bf27b3f8f86f51c",
            "packagetype": "bdist_wheel",
            "python_version": "py2.py3",
            "requires_python": null,
            "size": 64695,
            "upload_time": "2018-11-13T14:05:25",
            "url": "https://files.pythonhosted.org/packages/62/67/d2b73956ef790b9b42ab88776de99e8e046bceec6e1294fec94ea3c57f61/starttf-1.0.3-py2.py3-none-any.whl"
        }
    ]
}