CNN.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "class Conv3x3:\n",
    "  # A Convolution layer using 3x3 filters.\n",
    "\n",
    "  def __init__(self, num_filters): #kacta filter olacagı\n",
    "    self.num_filters = num_filters\n",
    "\n",
    "    # filters is a 3d array with dimensions (num_filters, 3, 3)\n",
    "    # We divide by 9 to reduce the variance of our initial values\n",
    "    self.filters = np.random.randn(num_filters, 3, 3) / 9 # 3*3 luk num filter kadar deger uretıyor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Conv3x3:\n",
    "  # ...\n",
    "\n",
    "  def iterate_regions(self, image):\n",
    "    '''\n",
    "    Generates all possible 3x3 image regions using valid padding.\n",
    "    - image is a 2d numpy array\n",
    "    '''\n",
    "    h, w = image.shape #resmin boyutlaru h ve w\n",
    "\n",
    "    for i in range(h - 2): # resimde son 2 satıra ilerlemiyor\n",
    "        for j in range(w - 2): # 3 e 3 oldugunda 2 verdik\n",
    "        im_region = image[i:(i + 3), j:(j + 3)]\n",
    "        yield im_region, i, j #herbir cagirimda kaldıgı yerden devam ediyor\n",
    "\n",
    "    def forward(self, input):\n",
    "    '''\n",
    "    Performs a forward pass of the conv layer using the given input.\n",
    "    Returns a 3d numpy array with dimensions (h, w, num_filters).\n",
    "    - input is a 2d numpy array\n",
    "    '''\n",
    "    h, w = input.shape\n",
    "    output = np.zeros((h - 2, w - 2, self.num_filters))\n",
    "\n",
    "    for im_region, i, j in self.iterate_regions(input):\n",
    "        output[i, j] = np.sum(im_region * self.filters, axis=(1, 2))\n",
    "\n",
    "    return output #resmin son hali"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import mnist #deneme\n",
    "from conv import Conv3x3\n",
    "\n",
    "# The mnist package handles the MNIST dataset for us!\n",
    "# Learn more at https://github.com/datapythonista/mnist\n",
    "train_images = mnist.train_images()\n",
    "train_labels = mnist.train_labels()\n",
    "\n",
    "conv = Conv3x3(8)\n",
    "output = conv.forward(train_images[0])\n",
    "print(output.shape) # (26, 26, 8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "class MaxPool2:\n",
    "  # A Max Pooling layer using a pool size of 2.\n",
    "\n",
    "  def iterate_regions(self, image):\n",
    "    '''\n",
    "    Generates non-overlapping 2x2 image regions to pool over.\n",
    "    - image is a 2d numpy array\n",
    "    '''\n",
    "    h, w, _ = image.shape\n",
    "    new_h = h // 2\n",
    "    new_w = w // 2\n",
    "\n",
    "    for i in range(new_h):\n",
    "        for j in range(new_w):\n",
    "        im_region = image[(i * 2):(i * 2 + 2), (j * 2):(j * 2 + 2)]\n",
    "        yield im_region, i, j\n",
    "\n",
    "    def forward(self, input):\n",
    "    '''\n",
    "    Performs a forward pass of the maxpool layer using the given input.\n",
    "    Returns a 3d numpy array with dimensions (h / 2, w / 2, num_filters).\n",
    "    - input is a 3d numpy array with dimensions (h, w, num_filters)\n",
    "    '''\n",
    "    h, w, num_filters = input.shape\n",
    "    output = np.zeros((h // 2, w // 2, num_filters))\n",
    "\n",
    "    for im_region, i, j in self.iterate_regions(input):\n",
    "      output[i, j] = np.amax(im_region, axis=(0, 1))#kendıne verilen matristekı max degerini alıyor\n",
    "\n",
    "    return output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "class Softmax:\n",
    "  # A standard fully-connected layer with softmax activation.\n",
    "\n",
    "  def __init__(self, input_len, nodes):\n",
    "    # We divide by input_len to reduce the variance of our initial values\n",
    "    self.weights = np.random.randn(input_len, nodes) / input_len\n",
    "    self.biases = np.zeros(nodes)\n",
    "\n",
    "    def forward(self, input):\n",
    "    '''\n",
    "    Performs a forward pass of the softmax layer using the given input.\n",
    "    Returns a 1d numpy array containing the respective probability values.\n",
    "    - input can be any array with any dimensions.\n",
    "    '''\n",
    "    input = input.flatten()\n",
    "\n",
    "    input_len, nodes = self.weights.shape\n",
    "\n",
    "    totals = np.dot(input, self.weights) + self.biases#skaler iç carpım \"dot\" inpuları al karsılıkları elamaları carp ve bunları topla tek bir deger uretir\n",
    "    exp = np.exp(totals)#\n",
    "    return exp / np.sum(exp, axis=0)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}