{
    "info": {
        "author": "Andrew Farley",
        "author_email": "amf7crazy@gmail.com",
        "bugtrack_url": null,
        "classifiers": [
            "License :: OSI Approved :: MIT License",
            "Operating System :: OS Independent",
            "Programming Language :: Python :: 3"
        ],
        "description": "# NeuralNetworkPY\nA basic, easy-to-use, neural network library built from scratch in python. Only dependency is numpy.\n\n## Examples\nI find examples are what I want when I go to a readme so I'm going to start with it. References for everything in this package can be found below.\n\n### Creating a Network to Understand XOR\n```python\n    #Initialize the network with 2 inputs, 1 hidden layer with 5 neurons, 2 outputs\n    nn = NeuralNetwork(2, [5], 2) \n\n    #Setting up the training data\n    data = []\n    data.append(np.array([0, 1]))\n    data.append(np.array([1, 0]))\n    data.append(np.array([1, 1]))\n    data.append(np.array([0, 0]))\n    out = [[0, 1], [0, 1], [1, 0], [1, 0]] #The expected output for every input added above \n    #(i.e. 0 xor 1 = 1  which in array form is [0, 1] because the guess function returns the index of the highest value, 1 in this case)\n\n    #Train the network\n    import random as rand\n    for i in range(50000):\n        index = rand.randint(0, 3) #randomizes which value we train with on each iteration\n        nn.train(data[index], out[index])\n\n    #Print the guess for every input\n    print('')\n    for ar in data:\n        print(str(ar[0]) + \" xor \" + str(ar[1]) + \" = \" + str(nn.guess(ar)))\n```\n\n### NeuralNetwork and MNIST Example (\"The hello world! of machine learning\")\n```python\n#This may take a long time to run, 2-10 minutes, depending on the speed of your cpu\n#I know it isn't commented the best, I've been writing documentation for hours at this point lol\n\nimport NeuralNetworkPY\nimport numpy as np\nimport struct\nimport random as rand\n\n#Data obtained from http://yann.lecun.com/exdb/mnist/. These files MUST be in same directory\n\n#Function obtained from https://gist.github.com/tylerneylon/ce60e8a06e7506ac45788443f7269e40\ndef read_idx(filename):\n    with open(filename, 'rb') as f:\n        zero, data_type, dims = struct.unpack('>HBB', f.read(4))\n        shape = tuple(struct.unpack('>I', f.read(4))[0] for d in range(dims))\n        return np.frombuffer(f.read(), dtype=np.uint8).reshape(shape)\n\ndef prepImgs(raw_images):\n    images = np.array([None for i in range(len(raw_images))])\n    for i in range(len(raw_images)):\n        images[i] = raw_images[i].flatten()\n        images[i] = images[i] / 255\n    return images\n\ndef createOutput(val):\n    out = []\n    for i in range(10):\n        if (i == val):\n            out.append(1.0)\n        else:\n            out.append(0.0)\n    return np.array([out]).transpose()\n\ndef guessToNum(guess):\n    max = guess[0]\n    index = 0\n    for i in range(len(guess)-1):\n        if (guess[i+1] > max):\n            max = guess[i+1]\n            index = i+1\n    return index\n\n#Load images and labels from files\nraw_images = read_idx('train-images.idx3-ubyte')\nlabels = read_idx('train-labels.idx1-ubyte')\n\n#Prep the images into a big array (need to normalize all values and flatten the 2D array to a 1D)\nimages = prepImgs(raw_images)\n\n#Setup network with 784 inputs (the images are 28*28 = 784), one hidden layer with 50 neurons, and 10 outputs (there are 10 possible values)\ndigitRecog = NeuralNetworkPY.NeuralNetwork(784, [50], 10)\ndigitRecog.setLearningRate(0.05)\n\nfor n in range(10): #train the network in 10 epochs\n    for i in range(len(images)): #in each epoch, train the network with a random image n times where n is the number of images\n        index = rand.randint(0, len(images)-1) #choose random image\n        digitRecog.train(images[index], createOutput(labels[index])) #train the network for one iteration\n\n#get and setup testing data\ntest_raw_images = read_idx('t10k-images.idx3-ubyte')\ntest_images = prepImgs(test_raw_images)\ntest_labels = read_idx('t10k-labels.idx1-ubyte')\ncorrect = 0\ntotal = len(test_images)\n\n#loop through all test images, count how many the network got right\nfor i in range(len(test_images)):\n    guess = digitRecog.feedForward(test_images[i])\n    if (guessToNum(guess) == test_labels[i]):\n        correct = correct + 1\n\n#print results\nprint(\"The Network guessed \" + str(correct) + \"/\" + str(total) + \" (\" + str(100*correct/total) + \"%) correct\")\n```\n\n\n## Object References\n### NeuralNetwork:\n  - Description: The NeuralNetwork object which is what is trained and where inputs are passed in. This is what you'll want\n  - Functions:\n    - `init(inputNum, hiddenNum, outputNum)` (the constructor):\n      - Description: Creates a NeuralNetwork object with the structure given as inputs\n      - Inputs:\n        - `inputNum`\n          - Description: The number of inputs to the network (size of input array)\n          - Type Required: int\n        - `hiddenNum`\n          - Description: The size of every hidden layer in the network\n          - Type Required: list of int\n        - `outputNum`\n          - Description: The number of outputs from the network (size of output array)\n          - Type Required: int\n    - `feedForward(inputData)`:\n      - Description: Computes the feed forwards algorithm for the network. Will propagate a given input array through every layer to produce an output array.\n      - Inputs:\n        - `inputData`\n          - Description: The input array to be propogated through the network.\n          - Type Required: 1D python list or numpy array (such as [1, 2, 3])\n      - Outputs:\n        - `probabilities`\n          - Description: The confidence values for every possible output.\n          - Type: Column vector in a numpy array. Will have shape (n, 1) where n is the number of elements.\n    - `train(inputData, expectedOutput)`:\n      - Description: Processes one iteration of network training, will adjust the weights and bias based on the error between the expected output and the current output for a given input.\n      - Inputs:\n        - `inputData`\n          - Description: The input array to be trained with\n          - Type Required: 1D python list or numpy array (such as [1, 2, 3])\n        - `expectedOutput`\n          - Description: What the given input array should produce in the network\n          - Type Required: 1D python list or numpy array (such as [1, 2, 3])\n    - `guess(inputData)`:\n      - Description: Produces a guess using forward propagation for a given input. The guess is the highest confidence value in the output array from feedForward()\n      - Inputs:\n        - `inputData`\n          - Description: The input array to get a guess from\n          - Type Required: 1D python list or numpy array (such as [1, 2, 3])\n      - Outputs:\n        - `output`\n          - Description: The index of the highest confidence value in the feedForward output.\n          - Type: int\n    - `setLearningRate(newRate)`:\n      - Description: Sets the learning rate of the network\n      - Inputs:\n        - `newRate`\n          - Description: The new learning rate\n          - Type Required: float\n    - `getLearningRate()`:\n      - Description: Gets the learning rate of the network\n      - Outputs:\n        - `output`\n          - Description: The learning rate of the current network.\n          - Type: float\n    - `save(fileName)`:\n      - Description: Saves the network to a text file for later use relative to the current directory. Format is provided below.\n      - Inputs:\n        - `fileName`\n          - Description: The name of the file (ex. \"test.txt\").\n          - Type Required: str\n    - `setLayers(weights, biases)`:\n      - Description: Sets the given network based on a list of weights and biases.\n      - Inputs:\n        - `weights`\n          - Description: A python list (must be the same size as self.layers) containing the new weight matricies for each layer.\n          - Type Required: Python list of 2D numpy arrays\n        - `biases`\n          - Description: A python list (must be the same size as self.layers) containing the new bias matricies for each layer.\n          - Type Required: Python list of 2D numpy arrays\n\n### Layer:\n  - Description: Used by the NeuralNetwork object for every layer (the inputs not being a layer). Contains information for every neuron in the layer which includes the weight connections and biases.\n  - Functions:\n    - `init(inputSize, selfSize)` (the constructor):\n      - Description: Creates a Layer object with the structure as given by parameters\n      - Inputs:\n        - `inputSize`\n          - Description: The number of inputs from the previous layer. Could also be thought of as the number of neurons on the previous layer.\n          - Type Required: int\n        - `selfSize`\n          - Description: The number of neurons on the current layer.\n          - Type Required: int\n    - `compute(inputData)`:\n      - Description: Propagates a given array of inputs through the current layer. This involves computing the necessary weighted sums for every neuron and adding the bias. Matrix math is used.\n      - Inputs:\n        - `inputData`\n          - Description: The input array to propagate through this layer\n          - Type Required: A column vector of inputs between -1 and 1. The shape of the numpy array is (n, 1) where n is the number of elements.\n      - Outputs:\n        - `out`\n          - Description: The output from the current layer with the given input.\n          - Type: A column vector of inputs between -1 and 1. The shape of the numpy array is (n, 1) where n is the number of elements.\n    - `stringify()`:\n      - Description: Gives a string representation of the layer. Used for saving the network to a text file.\n      - Outputs:\n        - `weightsStr`\n          - Description: The string of the weights matrix, stripped of newline characters.\n          - Type: str\n        - `biasStr`\n          - Description: The string of the bias matrix, stripped of newline characters.\n          - Type: str\n    - `setVals(weights, bias)`:\n      - Description: Sets the weights and bias of the current layer to the given inputs.\n      - Inputs:\n        - `weights`\n          - Description: A matrix of the weights, needs to match the size of the current one.\n          - Type Required: 2D numpy array\n        - `bias`\n          - Description: A matrix of the bias, needs to match the size of the current one.\n          - Type Required: 2D numpy array\n\n## Helper Functions\n### `checkReform(data)`\n  - Description: Checks an input data list or array, ensures it is a column vector, reshapes it if neccessary, returns the clean output.\n  - Inputs:\n    - `data`\n      - Description: A list or array that is going to be propagated through the array.\n      - Type Required: python list or numpy array\n  - Outputs:\n    - `out`\n      - Description: The cleaned input, is ready for use by a NeuralNetwork object\n      - Type: A column vector with the shape of the numpy array being (n, 1) where n is the number of elements.\n\n### `writeLine(file, text)`\n  - Description: Writes a line to an open file object with write permissions.\n  - Inputs:\n    - `file`\n      - Description: The file object to write to, must have write permissions.\n      - Type Required: _io.TextIOWrapper object\n    - `text`\n      - Description: The text to write to the line\n      - Type Required: str\n\n### `readLine(file)`\n  - Description: Reads the next line from an open file object with read permissions evaluated as a python variable.\n  - Inputs:\n    - `file`\n      - Description: The file object to read from, must have read permissions.\n      - Type Required: _io.TextIOWrapper object\n  - Outputs:\n    - `output`\n      - Description: Evaluates the line recieved as a python value. (i.e. the string \"array([1, 2, 3])\" would evaluate as a numpy array containing [1, 2, 3])\n      - Type: depends on the string evaluated\n\n### `loadNetwork(fileName)`\n  - Description: Reads a network info file with the given name and recreates a Neural Network out of it.\n  - Inputs:\n    - `fileName`\n      - Description: The name of the saved Neural Network data file. (i.e. \"test.txt\")\n      - Type Required: str\n  - Outputs:\n    - `out`\n      - Description: A Neural Network with the same specifications as outlined by the file data.\n      - Type: NeuralNetwork object\n\n\n## Save File Format\n**line**    **value**\n1. inputNum \n2. hiddenNum\n3. outputNum\n4. learningRate \n5. number of layers (layerNum) \n6. layer[0] Weights \n7. layer[0] Bias \n\n......... \n\n(5+2xlayerNum-1). layer[layerNum-1] Weights \n\n(5+2xlayerNum). layer[layerNum-1] Bias \n\n\n## Dev Notes\nI am currently working on building a Convolutional Neural Network from scratch modelled after the Neural Network created here. The code that is being worked on can be found here as well.\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/And1210/NeuralNetworkPY",
        "keywords": "",
        "license": "",
        "maintainer": "",
        "maintainer_email": "",
        "name": "NeuralNetworkPY",
        "package_url": "https://pypi.org/project/NeuralNetworkPY/",
        "platform": "",
        "project_url": "https://pypi.org/project/NeuralNetworkPY/",
        "project_urls": {
            "Homepage": "https://github.com/And1210/NeuralNetworkPY"
        },
        "release_url": "https://pypi.org/project/NeuralNetworkPY/0.0.2/",
        "requires_dist": null,
        "requires_python": "",
        "summary": "An easy-to-use Neural Network library I'm building from scratch.",
        "version": "0.0.2"
    },
    "last_serial": 4643185,
    "releases": {
        "0.0.1": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "3f6b11ed3543be56d781a67d16204087",
                    "sha256": "ed37762ec2a7b6fdf8ec8e57e8b3d811adf22c4aaa01c5f40a05c6f8c87473a6"
                },
                "downloads": -1,
                "filename": "NeuralNetworkPY-0.0.1-py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "3f6b11ed3543be56d781a67d16204087",
                "packagetype": "bdist_wheel",
                "python_version": "py3",
                "requires_python": null,
                "size": 5464,
                "upload_time": "2018-12-28T08:58:23",
                "url": "https://files.pythonhosted.org/packages/06/8d/a295c3cf602baca62f4ee7d4b87996102f96807e93a5f924793c46c21e60/NeuralNetworkPY-0.0.1-py3-none-any.whl"
            },
            {
                "comment_text": "",
                "digests": {
                    "md5": "e5552ac411b1c643e39573a7e2bb4902",
                    "sha256": "3f429f565a83cd28c56747a80689013b7414167cd70de60523c1fcf739ccc269"
                },
                "downloads": -1,
                "filename": "NeuralNetworkPY-0.0.1.tar.gz",
                "has_sig": false,
                "md5_digest": "e5552ac411b1c643e39573a7e2bb4902",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 4211,
                "upload_time": "2018-12-28T08:58:25",
                "url": "https://files.pythonhosted.org/packages/8b/90/aeef5ebb067c622c472afc3e97475c36b1f654dc39f08f6419f79682b762/NeuralNetworkPY-0.0.1.tar.gz"
            }
        ],
        "0.0.2": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "6bc7463c0f35c4e0b4943a428e7e78fd",
                    "sha256": "7dbab38c9b8db5c6bf5ccd4dd0479055a3ab232e58b5d7389f46085adbee423a"
                },
                "downloads": -1,
                "filename": "NeuralNetworkPY-0.0.2-py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "6bc7463c0f35c4e0b4943a428e7e78fd",
                "packagetype": "bdist_wheel",
                "python_version": "py3",
                "requires_python": null,
                "size": 8901,
                "upload_time": "2018-12-29T04:21:39",
                "url": "https://files.pythonhosted.org/packages/23/22/9b45954957b47232707c6f8cd9d91ac6db0513b13422166f169f05f34b7a/NeuralNetworkPY-0.0.2-py3-none-any.whl"
            },
            {
                "comment_text": "",
                "digests": {
                    "md5": "894bf8241f9f5806e51aaafca19e779f",
                    "sha256": "7e7e5e2b020d46f789bfb8a870d116e3dcb325881ae7d8efa8353244a3ca16e3"
                },
                "downloads": -1,
                "filename": "NeuralNetworkPY-0.0.2.tar.gz",
                "has_sig": false,
                "md5_digest": "894bf8241f9f5806e51aaafca19e779f",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 8202,
                "upload_time": "2018-12-29T04:21:41",
                "url": "https://files.pythonhosted.org/packages/d0/fd/548cf931ffddcc656ff23883fa2f4688a7e28a0fc5252f4943da4398b222/NeuralNetworkPY-0.0.2.tar.gz"
            }
        ]
    },
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "md5": "6bc7463c0f35c4e0b4943a428e7e78fd",
                "sha256": "7dbab38c9b8db5c6bf5ccd4dd0479055a3ab232e58b5d7389f46085adbee423a"
            },
            "downloads": -1,
            "filename": "NeuralNetworkPY-0.0.2-py3-none-any.whl",
            "has_sig": false,
            "md5_digest": "6bc7463c0f35c4e0b4943a428e7e78fd",
            "packagetype": "bdist_wheel",
            "python_version": "py3",
            "requires_python": null,
            "size": 8901,
            "upload_time": "2018-12-29T04:21:39",
            "url": "https://files.pythonhosted.org/packages/23/22/9b45954957b47232707c6f8cd9d91ac6db0513b13422166f169f05f34b7a/NeuralNetworkPY-0.0.2-py3-none-any.whl"
        },
        {
            "comment_text": "",
            "digests": {
                "md5": "894bf8241f9f5806e51aaafca19e779f",
                "sha256": "7e7e5e2b020d46f789bfb8a870d116e3dcb325881ae7d8efa8353244a3ca16e3"
            },
            "downloads": -1,
            "filename": "NeuralNetworkPY-0.0.2.tar.gz",
            "has_sig": false,
            "md5_digest": "894bf8241f9f5806e51aaafca19e779f",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": null,
            "size": 8202,
            "upload_time": "2018-12-29T04:21:41",
            "url": "https://files.pythonhosted.org/packages/d0/fd/548cf931ffddcc656ff23883fa2f4688a7e28a0fc5252f4943da4398b222/NeuralNetworkPY-0.0.2.tar.gz"
        }
    ]
}