Image predictor.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "cheap-price",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import h5py\n",
    "from collections import Counter\n",
    "import tensorflow as tf\n",
    "import keras"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "simple-resistance",
   "metadata": {},
   "outputs": [],
   "source": [
    "f = h5py.File('mouse1sample1.hdf5','r')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "biological-batman",
   "metadata": {},
   "outputs": [],
   "source": [
    "annotated_cells = []\n",
    "annotations = []\n",
    "for cell_id in f['cells']:\n",
    "    cell = f['cells'][cell_id]\n",
    "    ann = dict(cell.attrs)['annotation']\n",
    "    if ann != 'unannotated':\n",
    "        annotated_cells.append(cell_id)\n",
    "        annotations.append(ann)\n",
    "annotations_count = Counter(annotations)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "national-column",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of Annotated Samples: 18453\n",
      "Num of Cell Types: 93\n",
      "Max Number of Samples per Cell Type: 2136\n",
      "Min Number of Samples per Cell Type: 1\n",
      "Average Number of Samples per Cell Type: 198.41935483870967\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<BarContainer object of 93 artists>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"Number of Annotated Samples:\", len(annotated_cells))\n",
    "print(\"Num of Cell Types:\", len(annotations_count))\n",
    "print(\"Max Number of Samples per Cell Type:\", max(annotations_count.values()))\n",
    "print(\"Min Number of Samples per Cell Type:\", min(annotations_count.values()))\n",
    "print(\"Average Number of Samples per Cell Type:\", sum(annotations_count.values())/93)\n",
    "\n",
    "#graph of num of samples across cell-types \n",
    "plt.bar(np.arange(0,93), annotations_count.values())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "continued-firewall",
   "metadata": {},
   "outputs": [],
   "source": [
    "# creating one-hot coded vectors for cell-type (labels for multi)\n",
    "annotation_indices = {k: v for v, k in enumerate(list(annotations_count.keys()))}\n",
    "labels = np.zeros((len(annotated_cells), len(annotation_indices)))\n",
    "for index in range(0, len(annotated_cells)):\n",
    "    cell = f['cells'][annotated_cells[index]]\n",
    "    ann = dict(cell.attrs)['annotation']\n",
    "    arr_index = annotation_indices[ann]\n",
    "    labels[index][arr_index] = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "occasional-element",
   "metadata": {},
   "outputs": [],
   "source": [
    "# data = np.zeros(shape=(len(annotated_cells), 288, 432)).astype('uint8')\n",
    "\n",
    "# fig = plt.figure()\n",
    "# plt.gray()\n",
    "# idx = 0\n",
    "# for cell_id in annotated_cells:\n",
    "#     cell = f['cells'][cell_id]\n",
    "#     keys = list(cell['boundaries'].keys())\n",
    "#     midpoint = keys[int(len(keys)/2)]\n",
    "#     boundary = cell['boundaries'][midpoint]\n",
    "#     xs = boundary[:,0]\n",
    "#     ys = boundary[:,1]\n",
    "#     plt.plot(xs,ys)\n",
    "#     plt.axis('equal')\n",
    "#     plt.xticks([])\n",
    "#     plt.yticks([])\n",
    "#     fig.canvas.draw()\n",
    "#     fig_array = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)\n",
    "#     fig_array = fig_array.reshape(fig.canvas.get_width_height()[::-1] + (3,))\n",
    "#     data[idx] = fig_array[:,:,0].astype('uint8')\n",
    "#     idx += 1\n",
    "#     plt.clf()\n",
    "#     print(f\"\\r{idx}\", end=\"\")\n",
    "\n",
    "# np.save('boundary_arrays', data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e5741609",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = np.load('boundary_arrays.npy')\n",
    "data = np.repeat(data[..., np.newaxis], 3, -1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "electrical-costs",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(12917, 288, 432, 3) (12917, 93)\n",
      "(2768, 288, 432, 3) (2768, 93)\n",
      "(2768, 288, 432, 3) (2768, 93)\n"
     ]
    }
   ],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "train, test, train_y, test_y = train_test_split(data, labels, test_size=0.3)\n",
    "\n",
    "test, val, test_y, val_y = train_test_split(test, test_y, test_size=0.5)\n",
    "print(train.shape, train_y.shape)\n",
    "print(val.shape, val_y.shape)\n",
    "print(test.shape, test_y.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "quiet-reliance",
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "base_model = tf.keras.applications.resnet50.ResNet50(\n",
    "    include_top=False, weights='imagenet', input_tensor=None,\n",
    "    input_shape=(288,432, 3), pooling=None, classes=93\n",
    ")\n",
    "\n",
    "base_model.trainable=False\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "8d923ebe",
   "metadata": {},
   "outputs": [],
   "source": [
    "class CustomDataGen(tf.keras.utils.Sequence):\n",
    "    \n",
    "    def __init__(self, train, y,\n",
    "                 batch_size=16,\n",
    "                 input_size=(288, 432, 3),\n",
    "                 shuffle=True):\n",
    "        \n",
    "        self.train = train\n",
    "        self.y = y\n",
    "        self.batch_size = batch_size\n",
    "        self.input_size = input_size\n",
    "        self.shuffle = shuffle\n",
    "        \n",
    "        self.n = len(train)\n",
    "    \n",
    "    \n",
    "    def on_epoch_end(self):\n",
    "        pass\n",
    "    \n",
    "    def __getitem__(self, index):\n",
    "        x_batch = self.train[index * self.batch_size:(index + 1) * self.batch_size]\n",
    "        y_batch = self.y[index * self.batch_size:(index + 1) * self.batch_size]\n",
    "        return x_batch, y_batch\n",
    "    \n",
    "    def __len__(self):\n",
    "        return self.n // self.batch_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "3f3b787f",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_gen = CustomDataGen(train, train_y)\n",
    "val_gen = CustomDataGen(val, val_y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "be0710d1",
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = keras.Input(shape=(288, 432, 3))\n",
    "# We make sure that the base_model is running in inference mode here,\n",
    "# by passing `training=False`. This is important for fine-tuning, as you will\n",
    "# learn in a few paragraphs.\n",
    "x = base_model(inputs, training=False)\n",
    "# Convert features of shape `base_model.output_shape[1:]` to vectors\n",
    "x = keras.layers.GlobalAveragePooling2D()(x)\n",
    "# A Dense classifier with a single unit (binary classification)\n",
    "outputs = keras.layers.Dense(93)(x)\n",
    "model = keras.Model(inputs, outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1fa51f48",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "807/807 [==============================] - ETA: 0s - loss: 8.1953 - accuracy: 0.0042"
     ]
    }
   ],
   "source": [
    "lr= tf.keras.optimizers.schedules.ExponentialDecay(\n",
    "    0.0001,\n",
    "    decay_steps=100000,\n",
    "    decay_rate=0.95,\n",
    "    staircase=True)\n",
    "\n",
    "model.compile(optimizer=tf.optimizers.Adam(learning_rate=lr),\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "\n",
    "# model.fit(train, train_y, batch_size=1, epochs=20, validation_data=(val, val_y))\n",
    "history = model.fit(\n",
    "    train_gen, \n",
    "    epochs=10,\n",
    "    validation_data=val_gen\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "7480eb92",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-11-05 22:37:35.007694: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 1033150464 exceeds 10% of free system memory.\n",
      "2021-11-05 22:37:36.278561: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 1033150464 exceeds 10% of free system memory.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "87/87 [==============================] - 49s 513ms/step - loss: 7.6353 - accuracy: 0.1113\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[7.635293483734131, 0.1112716794013977]"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(test, test_y)"
   ]
  }
 ],
 "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": "3.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}