demo_omniglot.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "tf.logging.set_verbosity(tf.logging.ERROR)\n",
    "\n",
    "import tensorflow_datasets as tfds\n",
    "\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from skimage.transform import resize\n",
    "import random\n",
    "\n",
    "from pretrained_models.bisonai.model import OmniglotModelBisonai"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "### Parameters\n",
    "\n",
    "num_classes = 3\n",
    "learning_rate = 0.001\n",
    "epochs = 5\n",
    "checkpoint_path = \"pretrained_models/bisonai/1shot_3way_bisonai_ckpt_o15t/model.ckpt-99999\"\n",
    "\n",
    "def softmax(x):\n",
    "    e = np.exp(x)\n",
    "    return (e / np.sum(e, axis=1, keepdims=True))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data for fine-tuning & prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "sess = tf.Session()\n",
    "\n",
    "omniglot_train = tfds.load(name=\"omniglot\", split=tfds.Split.TRAIN).batch(2000)\n",
    "omniglot_example = omniglot_train.take(1)\n",
    "omniglot_example_iter = omniglot_example.make_initializable_iterator()\n",
    "sess.run(omniglot_example_iter.initializer)\n",
    "\n",
    "data = omniglot_example_iter.get_next()\n",
    "image = data['image']\n",
    "label = data['label']\n",
    "x, y = sess.run([image,label])\n",
    "\n",
    "X_train = resize(np.array(x[:num_classes,:,:,:])/255.0, (3, 28, 28, 1))\n",
    "y_train = np.array(y[:num_classes])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(3, 28, 28, 1)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([512, 854, 887])"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(X_train.shape)\n",
    "y_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x1440 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "### Showing Images of the 3 classes after reshape 28x28x1\n",
    "\n",
    "plt.figure(figsize=(20,20))\n",
    "plt.subplot(131)\n",
    "imgplot = plt.imshow(X_train[0,:,:,0], cmap=\"gist_gray\")\n",
    "plt.subplot(132)\n",
    "imgplot = plt.imshow(X_train[1,:,:,0], cmap=\"gist_gray\")\n",
    "plt.subplot(133)\n",
    "imgplot = plt.imshow(X_train[2,:,:,0], cmap=\"gist_gray\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "res = sum([np.where(y == c)[0].tolist() for c in y_train], [])\n",
    "[res.remove(i) for i in range(num_classes)]\n",
    "\n",
    "ind = random.choice(res)\n",
    "X_predict = resize(x[ind,:,:,0]/255.0, (28,28)).reshape((1, 28, 28,1))\n",
    "y_predict = y[ind]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "###### Showing the image with class to predict\n",
    "\n",
    "plt.figure(figsize=(6,6))\n",
    "imgplot = plt.imshow(X_predict[0,:,:,0], cmap=\"gist_gray\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preparing the model & prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "model = OmniglotModelBisonai(num_classes=3, **{'learning_rate':learning_rate})\n",
    "saver = tf.train.Saver()\n",
    "saver.restore(sess, checkpoint_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The predicted class number is 887 and the real class number is 887.\n",
      "The probability distribution is [[0.27328706 0.20178133 0.5249316 ]].\n"
     ]
    }
   ],
   "source": [
    "y_train_label = np.array([0, 1, 2])\n",
    "\n",
    "adam_var = [i for i in tf.all_variables() if \"Adam_1\" in i.name]\n",
    "\n",
    "for e in range(epochs):\n",
    "\n",
    "    #loss = sess.run(model.loss, feed_dict={model.input_ph: X_train.reshape(X_train.shape[:3]), model.label_ph: y_train_label})\n",
    "    #print(\"loss\", loss)\n",
    "\n",
    "    #logits = sess.run(model.logits, feed_dict={model.input_ph: X_predict.reshape(X_predict.shape[:3])})\n",
    "    #print(\"test softmax\", softmax(logits))\n",
    "    sess.run(model.minimize_op, feed_dict={model.input_ph: X_train.reshape(X_train.shape[:3]), model.label_ph:y_train_label})\n",
    "result, logits = sess.run([model.predictions, model.logits], feed_dict={model.input_ph: X_predict.reshape(X_predict.shape[:3])})\n",
    "\n",
    "print(\"The predicted class number is {} and the real class number is {}.\".format(y_train[result[0]],y_predict))\n",
    "print(\"The probability distribution is {}.\".format(softmax(logits)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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
}