Custom_Models_and_Training_with_TensorFlow.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Custom Models and Training with TensorFlow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/device:GPU:0'"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.test.gpu_device_name()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.config.experimental.list_physical_devices(device_type='GPU')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Using Tensorflow like NumPy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "t = tf.constant([1,2,3,4,5,6], shape=(2,3), dtype = 'float32')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(), dtype=int32, numpy=42>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.constant(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "TensorShape([2, 3])"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(3,), dtype=float32, numpy=array([4., 5., 6.], dtype=float32)>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(2, 2), dtype=float32, numpy=\n",
       "array([[2., 3.],\n",
       "       [5., 6.]], dtype=float32)>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t[:, 1:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(2, 1), dtype=float32, numpy=\n",
       "array([[2.],\n",
       "       [5.]], dtype=float32)>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t[..., 1, tf.newaxis]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(2, 3), dtype=float32, numpy=\n",
       "array([[11., 12., 13.],\n",
       "       [14., 15., 16.]], dtype=float32)>"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t + 10"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tf.square(t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(2, 2), dtype=float32, numpy=\n",
       "array([[14., 32.],\n",
       "       [32., 77.]], dtype=float32)>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t @ tf.transpose(t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(2, 3), dtype=float32, numpy=\n",
       "array([[  2.7182817,   7.389056 ,  20.085537 ],\n",
       "       [ 54.59815  , 148.41316  , 403.4288   ]], dtype=float32)>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.exp(t)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Some functions have a different name than in NumPy; for instance, tf.reduce_mean(), tf.reduce_sum(), tf.reduce_max(), and tf.math.log() are the equivalent of np.mean(), np.sum(), np.max() and np.log(). "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Using Keras.backend"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow import keras"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "K = keras.backend"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(2, 3), dtype=float32, numpy=\n",
       "array([[ 1.,  4.,  9.],\n",
       "       [16., 25., 36.]], dtype=float32)>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "K.square(t)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Tensors and NumPy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "a = np.array([2,4,6], dtype = np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dtype('float32')"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "a.dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(3,), dtype=float32, numpy=array([2., 4., 6.], dtype=float32)>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.constant(a)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "numpy.ndarray"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(t.numpy())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(3,), dtype=float32, numpy=array([ 4., 16., 36.], dtype=float32)>"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.square(a)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 1.,  4.,  9.],\n",
       "       [16., 25., 36.]], dtype=float32)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.square(t)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Type Conversions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Type conversions can significantly hurt performance, and they can easily go unnoticed when they are done automatically. To avoid this, TensorFlow does not perform any type conversions automatically: it just raises an exception if you try to execute an operation on tensors with incompatible types."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "ename": "InvalidArgumentError",
     "evalue": "cannot compute AddV2 as input #1(zero-based) was expected to be a float tensor but is a int32 tensor [Op:AddV2]",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mInvalidArgumentError\u001b[0m                      Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-22-1026a7e18cf4>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mconstant\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m2.\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mconstant\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m40\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py\u001b[0m in \u001b[0;36mbinary_op_wrapper\u001b[0;34m(x, y)\u001b[0m\n\u001b[1;32m    982\u001b[0m     \u001b[0;32mwith\u001b[0m \u001b[0mops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mname_scope\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;32mNone\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mop_name\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    983\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTensor\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTensor\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 984\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    985\u001b[0m       \u001b[0;32melif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msparse_tensor\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mSparseTensor\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    986\u001b[0m         \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py\u001b[0m in \u001b[0;36m_add_dispatch\u001b[0;34m(x, y, name)\u001b[0m\n\u001b[1;32m   1274\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mgen_math_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0madd\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1275\u001b[0m   \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1276\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mgen_math_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0madd_v2\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1277\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1278\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/tensorflow/python/ops/gen_math_ops.py\u001b[0m in \u001b[0;36madd_v2\u001b[0;34m(x, y, name)\u001b[0m\n\u001b[1;32m    478\u001b[0m         \u001b[0;32mpass\u001b[0m  \u001b[0;31m# Add nodes to the TensorFlow graph.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    479\u001b[0m     \u001b[0;32mexcept\u001b[0m \u001b[0m_core\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_NotOkStatusException\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 480\u001b[0;31m       \u001b[0m_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mraise_from_not_ok_status\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    481\u001b[0m   \u001b[0;31m# Add nodes to the TensorFlow graph.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    482\u001b[0m   _, _, _op, _outputs = _op_def_library._apply_op_helper(\n",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36mraise_from_not_ok_status\u001b[0;34m(e, name)\u001b[0m\n\u001b[1;32m   6651\u001b[0m   \u001b[0mmessage\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmessage\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;34m\" name: \"\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mname\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0mname\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32mNone\u001b[0m \u001b[0;32melse\u001b[0m \u001b[0;34m\"\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   6652\u001b[0m   \u001b[0;31m# pylint: disable=protected-access\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 6653\u001b[0;31m   \u001b[0msix\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mraise_from\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcore\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_status_to_exception\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcode\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmessage\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   6654\u001b[0m   \u001b[0;31m# pylint: enable=protected-access\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   6655\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/six.py\u001b[0m in \u001b[0;36mraise_from\u001b[0;34m(value, from_value)\u001b[0m\n",
      "\u001b[0;31mInvalidArgumentError\u001b[0m: cannot compute AddV2 as input #1(zero-based) was expected to be a float tensor but is a int32 tensor [Op:AddV2]"
     ]
    }
   ],
   "source": [
    "tf.constant(2.) + tf.constant(40)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "ename": "InvalidArgumentError",
     "evalue": "cannot compute AddV2 as input #1(zero-based) was expected to be a float tensor but is a double tensor [Op:AddV2]",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mInvalidArgumentError\u001b[0m                      Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-23-83093b0bc0ea>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mconstant\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m2.\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mconstant\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m2.\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdtype\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfloat64\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py\u001b[0m in \u001b[0;36mbinary_op_wrapper\u001b[0;34m(x, y)\u001b[0m\n\u001b[1;32m    982\u001b[0m     \u001b[0;32mwith\u001b[0m \u001b[0mops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mname_scope\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;32mNone\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mop_name\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    983\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTensor\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTensor\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 984\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    985\u001b[0m       \u001b[0;32melif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msparse_tensor\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mSparseTensor\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    986\u001b[0m         \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py\u001b[0m in \u001b[0;36m_add_dispatch\u001b[0;34m(x, y, name)\u001b[0m\n\u001b[1;32m   1274\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mgen_math_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0madd\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1275\u001b[0m   \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1276\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mgen_math_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0madd_v2\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1277\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1278\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/tensorflow/python/ops/gen_math_ops.py\u001b[0m in \u001b[0;36madd_v2\u001b[0;34m(x, y, name)\u001b[0m\n\u001b[1;32m    478\u001b[0m         \u001b[0;32mpass\u001b[0m  \u001b[0;31m# Add nodes to the TensorFlow graph.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    479\u001b[0m     \u001b[0;32mexcept\u001b[0m \u001b[0m_core\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_NotOkStatusException\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 480\u001b[0;31m       \u001b[0m_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mraise_from_not_ok_status\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    481\u001b[0m   \u001b[0;31m# Add nodes to the TensorFlow graph.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    482\u001b[0m   _, _, _op, _outputs = _op_def_library._apply_op_helper(\n",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36mraise_from_not_ok_status\u001b[0;34m(e, name)\u001b[0m\n\u001b[1;32m   6651\u001b[0m   \u001b[0mmessage\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmessage\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;34m\" name: \"\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mname\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0mname\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32mNone\u001b[0m \u001b[0;32melse\u001b[0m \u001b[0;34m\"\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   6652\u001b[0m   \u001b[0;31m# pylint: disable=protected-access\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 6653\u001b[0;31m   \u001b[0msix\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mraise_from\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcore\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_status_to_exception\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcode\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmessage\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   6654\u001b[0m   \u001b[0;31m# pylint: enable=protected-access\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   6655\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/media/shrey/Data/ModernTech/HOML/venv/lib/python3.6/site-packages/six.py\u001b[0m in \u001b[0;36mraise_from\u001b[0;34m(value, from_value)\u001b[0m\n",
      "\u001b[0;31mInvalidArgumentError\u001b[0m: cannot compute AddV2 as input #1(zero-based) was expected to be a float tensor but is a double tensor [Op:AddV2]"
     ]
    }
   ],
   "source": [
    "tf.constant(2.) + tf.constant(2., dtype=tf.float64)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(), dtype=float32, numpy=4.0>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.constant(2.) + tf.constant(2.)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Casting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(), dtype=float32, numpy=402.0>"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t2 = tf.constant(400, dtype=tf.float64)\n",
    "tf.constant(2.0) + tf.cast(t2, tf.float32)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "v = tf.Variable([[1., 2., 3.], [4., 5., 6.]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Variable 'UnreadVariable' shape=(2, 3) dtype=float32, numpy=\n",
       "array([[ 2.,  4.,  6.],\n",
       "       [ 8., 10., 12.]], dtype=float32)>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "v.assign(2*v)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Variable 'UnreadVariable' shape=(2, 3) dtype=float32, numpy=\n",
       "array([[ 2., 42.,  6.],\n",
       "       [ 8., 10., 12.]], dtype=float32)>"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "v[0, 1].assign(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Variable 'UnreadVariable' shape=(2, 3) dtype=float32, numpy=\n",
       "array([[ 2., 42.,  5.],\n",
       "       [ 8., 10.,  5.]], dtype=float32)>"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "v[:,2].assign([5, 5])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Variable 'UnreadVariable' shape=(2, 3) dtype=float32, numpy=\n",
       "array([[100.,  42.,   5.],\n",
       "       [  8.,  10., 200.]], dtype=float32)>"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "v.scatter_nd_update(indices=[[0,0],[1,2]], updates=[100., 200.])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Customizing Models and Training Algorithms"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Custom Loss Functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "def huber_fn(y_true, y_pred):\n",
    "    error = y_true - y_pred\n",
    "    is_small_error = tf.abs(error) < 1\n",
    "    squared_loss = tf.square(error) / 2\n",
    "    linear_loss = tf.abs(error) - .5\n",
    "    return tf.where(is_small_error, squared_loss, linear_loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.datasets import fetch_california_housing\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "housing = fetch_california_housing()\n",
    "X_train_full, X_test, y_train_full, y_test = train_test_split(\n",
    "    housing.data, housing.target.reshape(-1, 1), random_state=259)\n",
    "X_train, X_valid, y_train, y_valid = train_test_split(\n",
    "    X_train_full, y_train_full, random_state=259)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "scaler = StandardScaler()\n",
    "X_train_scaled = scaler.fit_transform(X_train)\n",
    "X_valid_scaled = scaler.transform(X_valid)\n",
    "X_test_scaled = scaler.transform(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x252 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(8, 3.5))\n",
    "z = np.linspace(-4, 4, 200)\n",
    "plt.plot(z, huber_fn(0, z), \"b-\", linewidth=2, label=\"huber($z$)\")\n",
    "plt.plot(z, z**2 / 2, \"b:\", linewidth=1, label=r\"$\\frac{1}{2}  z^2$\")\n",
    "plt.plot([-1, -1], [0, huber_fn(0., -1.)], \"r--\")\n",
    "plt.plot([1, 1], [0, huber_fn(0., 1.)], \"r--\")\n",
    "plt.gca().axhline(y=0, color='k')\n",
    "plt.gca().axvline(x=0, color='k')\n",
    "plt.axis([-4, 4, 0, 4])\n",
    "plt.grid(True)\n",
    "plt.xlabel(\"$z$\")\n",
    "plt.legend(fontsize=14)\n",
    "plt.title(\"Huber loss\", fontsize=14)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_shape = X_train.shape[1:]\n",
    "\n",
    "model = keras.models.Sequential([\n",
    "    keras.layers.Dense(30, activation=\"selu\", kernel_initializer=\"lecun_normal\",\n",
    "                       input_shape=input_shape),\n",
    "    keras.layers.Dense(1),\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=huber_fn, optimizer=\"nadam\", metrics=[\"mae\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.5791 - mae: 0.9354 - val_loss: 0.2444 - val_mae: 0.5480\n",
      "Epoch 2/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2161 - mae: 0.5106 - val_loss: 0.2321 - val_mae: 0.5292\n",
      "Epoch 3/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2075 - mae: 0.4975 - val_loss: 0.2244 - val_mae: 0.5182\n",
      "Epoch 4/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2024 - mae: 0.4904 - val_loss: 0.2192 - val_mae: 0.5091\n",
      "Epoch 5/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1980 - mae: 0.4837 - val_loss: 0.2132 - val_mae: 0.5046\n",
      "Epoch 6/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1940 - mae: 0.4785 - val_loss: 0.2104 - val_mae: 0.5004\n",
      "Epoch 7/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1911 - mae: 0.4746 - val_loss: 0.2066 - val_mae: 0.4937\n",
      "Epoch 8/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1881 - mae: 0.4706 - val_loss: 0.2012 - val_mae: 0.4870\n",
      "Epoch 9/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1858 - mae: 0.4679 - val_loss: 0.1978 - val_mae: 0.4852\n",
      "Epoch 10/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1838 - mae: 0.4657 - val_loss: 0.1994 - val_mae: 0.4851\n",
      "Epoch 11/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1829 - mae: 0.4631 - val_loss: 0.1947 - val_mae: 0.4805\n",
      "Epoch 12/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1820 - mae: 0.4613 - val_loss: 0.1961 - val_mae: 0.4853\n",
      "Epoch 13/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1799 - mae: 0.4598 - val_loss: 0.1932 - val_mae: 0.4774\n",
      "Epoch 14/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1799 - mae: 0.4592 - val_loss: 0.1943 - val_mae: 0.4784\n",
      "Epoch 15/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1785 - mae: 0.4565 - val_loss: 0.1922 - val_mae: 0.4760\n",
      "Epoch 16/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1782 - mae: 0.4552 - val_loss: 0.1907 - val_mae: 0.4748\n",
      "Epoch 17/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1768 - mae: 0.4530 - val_loss: 0.1906 - val_mae: 0.4792\n",
      "Epoch 18/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1761 - mae: 0.4531 - val_loss: 0.1903 - val_mae: 0.4724\n",
      "Epoch 19/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1755 - mae: 0.4508 - val_loss: 0.1908 - val_mae: 0.4741\n",
      "Epoch 20/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1748 - mae: 0.4496 - val_loss: 0.1886 - val_mae: 0.4724\n",
      "Epoch 21/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1737 - mae: 0.4476 - val_loss: 0.1860 - val_mae: 0.4677\n",
      "Epoch 22/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1726 - mae: 0.4457 - val_loss: 0.1862 - val_mae: 0.4679\n",
      "Epoch 23/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1726 - mae: 0.4454 - val_loss: 0.1861 - val_mae: 0.4687\n",
      "Epoch 24/30\n",
      "363/363 [==============================] - 1s 4ms/step - loss: 0.1717 - mae: 0.4448 - val_loss: 0.1855 - val_mae: 0.4654\n",
      "Epoch 25/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1711 - mae: 0.4430 - val_loss: 0.1839 - val_mae: 0.4655\n",
      "Epoch 26/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1704 - mae: 0.4424 - val_loss: 0.1850 - val_mae: 0.4619\n",
      "Epoch 27/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1699 - mae: 0.4408 - val_loss: 0.1828 - val_mae: 0.4606\n",
      "Epoch 28/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1698 - mae: 0.4408 - val_loss: 0.1817 - val_mae: 0.4619\n",
      "Epoch 29/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1688 - mae: 0.4391 - val_loss: 0.1815 - val_mae: 0.4623\n",
      "Epoch 30/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1689 - mae: 0.4389 - val_loss: 0.1838 - val_mae: 0.4605\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8e3333160>"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled, y_train, epochs=30,\n",
    "          validation_data=(X_valid_scaled, y_valid))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Saving and Loading Models That Contain Custom Components"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save('model_withh_custom_loss.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.load_model('model_withh_custom_loss.h5', custom_objects={'huber_fn': huber_fn})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1680 - mean_absolute_error: 0.4376 - val_loss: 0.1825 - val_mean_absolute_error: 0.4591\n",
      "Epoch 2/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1672 - mean_absolute_error: 0.4363 - val_loss: 0.1818 - val_mean_absolute_error: 0.4607\n",
      "Epoch 3/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1669 - mean_absolute_error: 0.4363 - val_loss: 0.1831 - val_mean_absolute_error: 0.4656\n",
      "Epoch 4/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1664 - mean_absolute_error: 0.4354 - val_loss: 0.1823 - val_mean_absolute_error: 0.4591\n",
      "Epoch 5/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1664 - mean_absolute_error: 0.4351 - val_loss: 0.1793 - val_mean_absolute_error: 0.4600\n",
      "Epoch 6/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1657 - mean_absolute_error: 0.4336 - val_loss: 0.1803 - val_mean_absolute_error: 0.4581\n",
      "Epoch 7/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1657 - mean_absolute_error: 0.4341 - val_loss: 0.1813 - val_mean_absolute_error: 0.4578\n",
      "Epoch 8/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1653 - mean_absolute_error: 0.4332 - val_loss: 0.1788 - val_mean_absolute_error: 0.4526\n",
      "Epoch 9/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1651 - mean_absolute_error: 0.4330 - val_loss: 0.1791 - val_mean_absolute_error: 0.4544\n",
      "Epoch 10/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1643 - mean_absolute_error: 0.4308 - val_loss: 0.1786 - val_mean_absolute_error: 0.4576\n",
      "Epoch 11/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1641 - mean_absolute_error: 0.4313 - val_loss: 0.1784 - val_mean_absolute_error: 0.4531\n",
      "Epoch 12/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1638 - mean_absolute_error: 0.4308 - val_loss: 0.1805 - val_mean_absolute_error: 0.4547\n",
      "Epoch 13/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1635 - mean_absolute_error: 0.4305 - val_loss: 0.1778 - val_mean_absolute_error: 0.4521\n",
      "Epoch 14/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1637 - mean_absolute_error: 0.4308 - val_loss: 0.1779 - val_mean_absolute_error: 0.4504\n",
      "Epoch 15/15\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1627 - mean_absolute_error: 0.4289 - val_loss: 0.1771 - val_mean_absolute_error: 0.4503\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8e0171c18>"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled, y_train, epochs = 15, validation_data=(X_valid_scaled, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_huber(threshold = 1.0):\n",
    "    def huber_fn(y_true, y_pred):\n",
    "        error = y_true - y_pred\n",
    "        is_small_error = tf.abs(error) < threshold\n",
    "        squared_loss = tf.square(error) / 2\n",
    "        linear_loss  = threshold * tf.abs(error) - threshold**2 / 2\n",
    "        return tf.where(is_small_error, squared_loss, linear_loss)\n",
    "    return huber_fn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=create_huber(2.0), optimizer=\"nadam\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1822 - val_loss: 0.2020\n",
      "Epoch 2/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1821 - val_loss: 0.1992\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8d3ffba20>"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled, y_train, epochs=2,\n",
    "          validation_data=(X_valid_scaled, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save(\"my_model_with_a_custom_loss_threshold_2.h5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.load_model(\"my_model_with_a_custom_loss_threshold_2.h5\",\n",
    "                                custom_objects={\"huber_fn\": create_huber(2.0)})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1812 - val_loss: 0.2010\n",
      "Epoch 2/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1828 - val_loss: 0.1988\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8d2f210f0>"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\n",
    "model.fit(X_train_scaled, y_train, epochs=2,\n",
    "          validation_data=(X_valid_scaled, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "class HuberLoss(keras.losses.Loss):\n",
    "    def __init__(self, threshold=1.0, **kwargs):\n",
    "        self.threshold = threshold\n",
    "        super().__init__(**kwargs)\n",
    "    def call(self, y_true, y_pred):\n",
    "        error = y_true - y_pred\n",
    "        is_small_error = tf.abs(error) < self.threshold\n",
    "        squared_loss = tf.square(error) / 2\n",
    "        linear_loss  = self.threshold * tf.abs(error) - self.threshold**2 / 2\n",
    "        return tf.where(is_small_error, squared_loss, linear_loss)\n",
    "    def get_config(self):\n",
    "        base_config = super().get_config()\n",
    "        return {**base_config, \"threshold\": self.threshold}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.Dense(30, activation=\"selu\", kernel_initializer=\"lecun_normal\",\n",
    "                       input_shape=input_shape),\n",
    "    keras.layers.Dense(1),\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=HuberLoss(2.), optimizer=\"nadam\", metrics=[\"mae\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.8260 - mae: 0.9715 - val_loss: 0.2857 - val_mae: 0.5600\n",
      "Epoch 2/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2479 - mae: 0.5164 - val_loss: 0.2508 - val_mae: 0.5213\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8d2e0f2b0>"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled, y_train, epochs=2,\n",
    "          validation_data=(X_valid_scaled, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save(\"my_model_with_a_custom_loss_class.h5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.load_model(\"my_model_with_a_custom_loss_class.h5\",\n",
    "                               custom_objects={\"HuberLoss\": HuberLoss})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2302 - mean_absolute_error: 0.4977 - val_loss: 0.2415 - val_mean_absolute_error: 0.5106\n",
      "Epoch 2/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2228 - mean_absolute_error: 0.4900 - val_loss: 0.2418 - val_mean_absolute_error: 0.5010\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8d2d254a8>"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled, y_train, epochs=2,\n",
    "          validation_data=(X_valid_scaled, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2.0"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.loss.threshold"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Custom Activation Functions, Initializers, Regularizers, and Constraints"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here are examples of a custom activation function (equivalent to keras.activations.softplus() or tf.nn.softplus()), a custom Glorot initializer (equivalent to keras.initializers.glorot_normal()), a custom ℓ1 regularizer (equivalent to keras.regularizers.l1(0.01)), and a custom constraint that ensures weights are all positive (equivalent to keras.constraints.nonneg() or tf.nn.relu())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "def my_softplus(z):\n",
    "    return tf.math.log(tf.exp(z) + 1.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "def my_glorot_initializer(shape, dtype = tf.float32):\n",
    "    stddev = tf.sqrt(2. / shape[0] + shape[1])\n",
    "    return tf.random.normal(shape, stddev = stddev, dtype=dtype)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "def my_l1_regularizer(weights):\n",
    "    return tf.reduce_sum(tf.abs(0.01 * weights))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [],
   "source": [
    "def my_positive_weights(weights):\n",
    "    return tf.where(weights < 0. , tf.zeros_like(weights), weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "layer = keras.layers.Dense(30, activation=my_softplus,\n",
    "                           kernel_initializer=my_glorot_initializer,\n",
    "                           kernel_regularizer=my_l1_regularizer,\n",
    "                           kernel_constraint=my_positive_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "keras.backend.clear_session()\n",
    "np.random.seed(259)\n",
    "tf.random.set_seed(259)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.Dense(30, activation=\"selu\", kernel_initializer=\"lecun_normal\",\n",
    "                       input_shape=input_shape),\n",
    "    keras.layers.Dense(1, activation=my_softplus,\n",
    "                       kernel_regularizer=my_l1_regularizer,\n",
    "                       kernel_constraint=my_positive_weights,\n",
    "                       kernel_initializer=my_glorot_initializer),\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=\"mse\", optimizer=\"nadam\", metrics=[\"mae\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "363/363 [==============================] - 1s 4ms/step - loss: nan - mae: nan - val_loss: nan - val_mae: nan\n",
      "Epoch 2/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: nan - mae: nan - val_loss: nan - val_mae: nan\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8e2178e10>"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled, y_train, epochs=2,\n",
    "          validation_data=(X_valid_scaled, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save(\"my_model_with_many_custom_parts.h5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.load_model(\n",
    "    \"my_model_with_many_custom_parts.h5\",\n",
    "    custom_objects={\n",
    "       \"my_l1_regularizer\": my_l1_regularizer,\n",
    "       \"my_positive_weights\": my_positive_weights,\n",
    "       \"my_glorot_initializer\": my_glorot_initializer,\n",
    "       \"my_softplus\": my_softplus,\n",
    "    })"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MyL1Regularizer(keras.regularizers.Regularizer):\n",
    "    def __init__(self, factor):\n",
    "        self.factor = factor\n",
    "    def __call__(self, weights):\n",
    "        return tf.reduce_sum(tf.abs(self.factor * weights))\n",
    "    def get_config(self):\n",
    "        return {'factor': self.factor}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "keras.backend.clear_session()\n",
    "np.random.seed(259)\n",
    "tf.random.set_seed(259)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.Dense(30, activation=\"selu\", kernel_initializer=\"lecun_normal\",\n",
    "                       input_shape=input_shape),\n",
    "    keras.layers.Dense(1, activation=my_softplus,\n",
    "                       kernel_regularizer=MyL1Regularizer(0.01),\n",
    "                       kernel_constraint=my_positive_weights,\n",
    "                       kernel_initializer=my_glorot_initializer),\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=\"mse\", optimizer=\"nadam\", metrics=[\"mae\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "363/363 [==============================] - 1s 4ms/step - loss: nan - mae: nan - val_loss: nan - val_mae: nan\n",
      "Epoch 2/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: nan - mae: nan - val_loss: nan - val_mae: nan\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8e39cbcf8>"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled, y_train, epochs=2,\n",
    "          validation_data=(X_valid_scaled, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save(\"my_model_with_many_custom_parts.h5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.load_model(\n",
    "    \"my_model_with_many_custom_parts.h5\",\n",
    "    custom_objects={\n",
    "       \"MyL1Regularizer\": MyL1Regularizer,\n",
    "       \"my_positive_weights\": my_positive_weights,\n",
    "       \"my_glorot_initializer\": my_glorot_initializer,\n",
    "       \"my_softplus\": my_softplus,\n",
    "    })"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Custom Metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "keras.backend.clear_session()\n",
    "np.random.seed(259)\n",
    "tf.random.set_seed(259)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.Dense(30, activation='selu', kernel_initializer = 'lecun_normal', input_shape = input_shape),\n",
    "    keras.layers.Dense(1),\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss = 'mse', optimizer='nadam', metrics = [create_huber(2.0)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 2.4209 - huber_fn: 0.8056\n",
      "Epoch 2/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.9679 - huber_fn: 0.2658\n",
      "Epoch 3/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.7834 - huber_fn: 0.2480\n",
      "Epoch 4/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.6497 - huber_fn: 0.2390\n",
      "Epoch 5/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.5727 - huber_fn: 0.2331\n",
      "Epoch 6/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.5062 - huber_fn: 0.2275\n",
      "Epoch 7/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4711 - huber_fn: 0.2223\n",
      "Epoch 8/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4470 - huber_fn: 0.2173\n",
      "Epoch 9/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4356 - huber_fn: 0.2141\n",
      "Epoch 10/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4280 - huber_fn: 0.2109\n",
      "Epoch 11/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4232 - huber_fn: 0.2088\n",
      "Epoch 12/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4206 - huber_fn: 0.2074\n",
      "Epoch 13/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4171 - huber_fn: 0.2056\n",
      "Epoch 14/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4153 - huber_fn: 0.2047\n",
      "Epoch 15/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4136 - huber_fn: 0.2039\n",
      "Epoch 16/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4108 - huber_fn: 0.2026\n",
      "Epoch 17/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4096 - huber_fn: 0.2019\n",
      "Epoch 18/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4071 - huber_fn: 0.2008\n",
      "Epoch 19/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4056 - huber_fn: 0.2001\n",
      "Epoch 20/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4043 - huber_fn: 0.1993\n",
      "Epoch 21/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4047 - huber_fn: 0.1991\n",
      "Epoch 22/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4093 - huber_fn: 0.1988\n",
      "Epoch 23/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4051 - huber_fn: 0.1981\n",
      "Epoch 24/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.4008 - huber_fn: 0.1970\n",
      "Epoch 25/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.3965 - huber_fn: 0.1954\n",
      "Epoch 26/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.3960 - huber_fn: 0.1950\n",
      "Epoch 27/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.3950 - huber_fn: 0.1943\n",
      "Epoch 28/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.3933 - huber_fn: 0.1935\n",
      "Epoch 29/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.3941 - huber_fn: 0.1934\n",
      "Epoch 30/30\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.3947 - huber_fn: 0.1932\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc8d2e00cc0>"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled, y_train, epochs = 30)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.69021991 0.1008456  0.26613551 ... 0.72405129 0.00300302 0.31897148]\n",
      "Epoch 1/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.0969 - huber_fn: 0.1917\n",
      "Epoch 2/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.0963 - huber_fn: 0.1911\n"
     ]
    }
   ],
   "source": [
    "model.compile(loss=create_huber(2.0), optimizer=\"nadam\", metrics=[create_huber(2.0)])\n",
    "sample_weight = np.random.rand(len(y_train))\n",
    "print(sample_weight)\n",
    "history = model.fit(X_train_scaled, y_train, epochs=2, sample_weight=sample_weight)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.09694576263427734, 0.09569035912599493)"
      ]
     },
     "execution_count": 79,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "history.history[\"loss\"][0], history.history[\"huber_fn\"][0] * sample_weight.mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Streaming metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(), dtype=float32, numpy=0.8>"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "precision = keras.metrics.Precision()\n",
    "precision([0,1,1,1,0,1,0,1],[1,1,0,1,0,1,0,1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(), dtype=float32, numpy=0.5>"
      ]
     },
     "execution_count": 81,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "precision([0, 1, 0, 0, 1, 0, 1, 1], [1, 0, 1, 1, 0, 0, 0, 0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(), dtype=float32, numpy=0.5>"
      ]
     },
     "execution_count": 82,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "precision.result()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<tf.Variable 'true_positives:0' shape=(1,) dtype=float32, numpy=array([4.], dtype=float32)>,\n",
       " <tf.Variable 'false_positives:0' shape=(1,) dtype=float32, numpy=array([4.], dtype=float32)>]"
      ]
     },
     "execution_count": 83,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "precision.variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "precision.reset_states()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [],
   "source": [
    "class HuberMetric(keras.metrics.Metric):\n",
    "    def __init__(self, threshold=1.0, **kwargs):\n",
    "        super().__init__(**kwargs) # handles base args (e.g., dtype)\n",
    "        self.threshold = threshold\n",
    "        self.huber_fn = create_huber(threshold)\n",
    "        self.total = self.add_weight(\"total\", initializer=\"zeros\")\n",
    "        self.count = self.add_weight(\"count\", initializer=\"zeros\")\n",
    "    def update_state(self, y_true, y_pred, sample_weight=None):\n",
    "        metric = self.huber_fn(y_true, y_pred)\n",
    "        self.total.assign_add(tf.reduce_sum(metric))\n",
    "        self.count.assign_add(tf.cast(tf.size(y_true), tf.float32))\n",
    "    def result(self):\n",
    "        return self.total / self.count\n",
    "    def get_config(self):\n",
    "        base_config = super().get_config()\n",
    "        return {**base_config, \"threshold\": self.threshold}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [],
   "source": [
    "m = HuberMetric(2.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(), dtype=float32, numpy=14.0>"
      ]
     },
     "execution_count": 87,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "m(tf.constant([[2.]]), tf.constant([[10.]]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(), dtype=float32, numpy=7.0>"
      ]
     },
     "execution_count": 88,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "m(tf.constant([[0.], [5.]]), tf.constant([[1.], [9.25]]))\n",
    "\n",
    "m.result()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<tf.Variable 'total:0' shape=() dtype=float32, numpy=21.0>,\n",
       " <tf.Variable 'count:0' shape=() dtype=float32, numpy=3.0>]"
      ]
     },
     "execution_count": 89,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "m.variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<tf.Variable 'total:0' shape=() dtype=float32, numpy=0.0>,\n",
       " <tf.Variable 'count:0' shape=() dtype=float32, numpy=0.0>]"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "m.reset_states()\n",
    "m.variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [],
   "source": [
    "keras.backend.clear_session()\n",
    "np.random.seed(259)\n",
    "tf.random.set_seed(259)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.Dense(30, activation=\"selu\", kernel_initializer=\"lecun_normal\",\n",
    "                       input_shape=input_shape),\n",
    "    keras.layers.Dense(1),\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=create_huber(2.0), optimizer=\"nadam\", metrics=[HuberMetric(2.0)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.7631 - huber_metric: 0.7631\n",
      "Epoch 2/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2616 - huber_metric: 0.2616\n",
      "Epoch 3/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2475 - huber_metric: 0.2475\n",
      "Epoch 4/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2381 - huber_metric: 0.2381\n",
      "Epoch 5/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2334 - huber_metric: 0.2334\n",
      "Epoch 6/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2265 - huber_metric: 0.2265\n",
      "Epoch 7/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2216 - huber_metric: 0.2216\n",
      "Epoch 8/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2155 - huber_metric: 0.2155\n",
      "Epoch 9/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2122 - huber_metric: 0.2122\n",
      "Epoch 10/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2068 - huber_metric: 0.2068\n",
      "Epoch 11/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2063 - huber_metric: 0.2063\n",
      "Epoch 12/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2044 - huber_metric: 0.2044\n",
      "Epoch 13/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2038 - huber_metric: 0.2038\n",
      "Epoch 14/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2021 - huber_metric: 0.2021\n",
      "Epoch 15/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.2017 - huber_metric: 0.2017\n",
      "Epoch 16/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1999 - huber_metric: 0.1999\n",
      "Epoch 17/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1983 - huber_metric: 0.1983\n",
      "Epoch 18/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1974 - huber_metric: 0.1974\n",
      "Epoch 19/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1968 - huber_metric: 0.1968\n",
      "Epoch 20/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1957 - huber_metric: 0.1957\n",
      "Epoch 21/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1956 - huber_metric: 0.1956\n",
      "Epoch 22/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1936 - huber_metric: 0.1936\n",
      "Epoch 23/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1941 - huber_metric: 0.1941\n",
      "Epoch 24/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1925 - huber_metric: 0.1925\n",
      "Epoch 25/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1921 - huber_metric: 0.1921\n",
      "Epoch 26/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1915 - huber_metric: 0.1915\n",
      "Epoch 27/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1911 - huber_metric: 0.1911\n",
      "Epoch 28/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1893 - huber_metric: 0.1893\n",
      "Epoch 29/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1902 - huber_metric: 0.1902\n",
      "Epoch 30/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1894 - huber_metric: 0.1894\n",
      "Epoch 31/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1888 - huber_metric: 0.1888\n",
      "Epoch 32/32\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.1882 - huber_metric: 0.1882\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7fc9a7f22438>"
      ]
     },
     "execution_count": 94,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_scaled.astype(np.float32), y_train.astype(np.float32), epochs=32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(11610, 8)"
      ]
     },
     "execution_count": 95,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train_scaled.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Custom Layers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The exponential layer is sometimes used in the output layer of a regression model when the values to predict have very different scales"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [],
   "source": [
    "exponential_layer = keras.layers.Lambda(lambda x: tf.exp(x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(3,), dtype=float32, numpy=array([0.36787945, 1.        , 2.7182817 ], dtype=float32)>"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "exponential_layer([-1., 0., 1.])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MyDense(keras.layers.Layer):\n",
    "    def __init__(self, units, activation = None, **kwargs):\n",
    "        super().__init__(**kwargs)\n",
    "        self.units = units\n",
    "        self.activation = keras.activations.get(activation)\n",
    "        \n",
    "    def build(self, batch_input_shape):\n",
    "        self.kernel = self.add_weight(\n",
    "            name = 'kernel', shape=[batch_input_shape[-1], self.units],\n",
    "            initializer='glorot_normal')\n",
    "        self.bias = self.add_weight(\n",
    "            name = 'bias', shape = [self.units], initializer='zeros')\n",
    "        super().build(batch_input_shape)\n",
    "    \n",
    "    def call(self, X):\n",
    "        return self.activation(X @ self.kernel + self.bias)\n",
    "    \n",
    "    def compute_output_shape(self, batch_input_shape):\n",
    "        return tf.TensorShape(batch_input_shape.as_list()[:-1] + [self.units])\n",
    "    \n",
    "    def get_config(self):\n",
    "        base_config = super().get_config()\n",
    "        return {**base_config, 'units': self.units,\n",
    "               'activation': keras.activations.serialize(self.activation)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [],
   "source": [
    "keras.backend.clear_session()\n",
    "np.random.seed(259)\n",
    "tf.random.set_seed(259)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.Sequential([\n",
    "    MyDense(30, activation=\"relu\", input_shape=input_shape),\n",
    "    MyDense(1)\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 2.8486 - val_loss: 0.7257\n",
      "Epoch 2/2\n",
      "363/363 [==============================] - 1s 3ms/step - loss: 0.5939 - val_loss: 0.5766\n",
      "162/162 [==============================] - 0s 1ms/step - loss: 0.5453\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.5453079342842102"
      ]
     },
     "execution_count": 102,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.compile(loss=\"mse\", optimizer=\"nadam\")\n",
    "model.fit(X_train_scaled, y_train, epochs=2,\n",
    "          validation_data=(X_valid_scaled, y_valid))\n",
    "model.evaluate(X_test_scaled, y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save(\"my_model_with_a_custom_layer.h5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.load_model(\"my_model_with_a_custom_layer.h5\",\n",
    "                                custom_objects={\"MyDense\": MyDense})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [],
   "source": [
    "keras.backend.clear_session()\n",
    "np.random.seed(259)\n",
    "tf.random.set_seed(259)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Custom Model"
   ]
  },
  {
   "attachments": {
    "d12b7106-9fc8-4b49-8667-25a17f8912b6.png": {
     "image/png": "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"
    }
   },
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![image.png](attachment:d12b7106-9fc8-4b49-8667-25a17f8912b6.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Suppose you want to build the model represented in figure"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To implement this model, it is best to first create a ResidualBlock layer, since we are going to create a couple of identical blocks (and we might want to reuse it in another model):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResidualBlock(keras.layers.Layer):\n",
    "    def __init__(self, n_layers, n_neurons, **kwargs):\n",
    "        super().__init__(**kwargs)\n",
    "        self.hidden = [keras.layers.Dense(n_neurons, activation = 'elu', kernel_initializer='he_normal') \n",
    "                          for _ in range(n_layers)]\n",
    "        \n",
    "    def call(self, inputs):\n",
    "        Z = inputs\n",
    "        for layer in self.hidden:\n",
    "            Z = layer(Z)\n",
    "            \n",
    "        return inputs + Z\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, let’s use the Subclassing API to define the model itself:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResidualRegressor(keras.Model):\n",
    "    def __init__(self, output_dim, **kwargs):\n",
    "        super().__init__(**kwargs)\n",
    "        self.hidden1 = keras.layers.Dense(30, activation = 'elu',\n",
    "                                         kernel_initializer = 'he_normal')\n",
    "        self.block1 = ResidualBlock(2,30)\n",
    "        self.block2 = ResidualBlock(2,30)\n",
    "        self.out = keras.layers.Dense(output_dim)\n",
    "        \n",
    "    def call(self, inputs):\n",
    "        Z = self.hidden1(inputs)\n",
    "        for _ in range(1 + 3):\n",
    "            Z = self.block1(Z)\n",
    "        Z = self.block2(Z)\n",
    "        return self.out(Z)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If you also want to be able to save the model using the save() method and load it using the keras.models.load_model() function, you must implement the get_config() method (as we did earlier) in both the ResidualBlock class and the ResidualRegressor class."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [],
   "source": [
    "keras.backend.clear_session()\n",
    "np.random.seed(259)\n",
    "tf.random.set_seed(259)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 31.6637 - val_loss: 11.8559\n",
      "Epoch 2/40\n",
      "363/363 [==============================] - 2s 4ms/step - loss: 4.8879 - val_loss: 12.6859\n",
      "Epoch 3/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 12.5674 - val_loss: 10.2993\n",
      "Epoch 4/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 9.6330 - val_loss: 4.3503\n",
      "Epoch 5/40\n",
      "363/363 [==============================] - 2s 6ms/step - loss: 2.8625 - val_loss: 14.2352\n",
      "Epoch 6/40\n",
      "363/363 [==============================] - 2s 6ms/step - loss: 8.6314 - val_loss: 0.8570\n",
      "Epoch 7/40\n",
      "363/363 [==============================] - 2s 6ms/step - loss: 0.7252 - val_loss: 0.7471\n",
      "Epoch 8/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.6578 - val_loss: 0.7654\n",
      "Epoch 9/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.6616 - val_loss: 0.5979\n",
      "Epoch 10/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.5243 - val_loss: 0.5680\n",
      "Epoch 11/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.4926 - val_loss: 0.5305\n",
      "Epoch 12/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.4647 - val_loss: 0.5318\n",
      "Epoch 13/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.4833 - val_loss: 0.5123\n",
      "Epoch 14/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.4612 - val_loss: 0.7142\n",
      "Epoch 15/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.5551 - val_loss: 0.8126\n",
      "Epoch 16/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.6778 - val_loss: 1.1061\n",
      "Epoch 17/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.7547 - val_loss: 1.3046\n",
      "Epoch 18/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 1.0724 - val_loss: 0.6979\n",
      "Epoch 19/40\n",
      "363/363 [==============================] - 2s 6ms/step - loss: 0.6568 - val_loss: 1.6878\n",
      "Epoch 20/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.9010 - val_loss: 2.1480\n",
      "Epoch 21/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.9538 - val_loss: 2.4545\n",
      "Epoch 22/40\n",
      "363/363 [==============================] - 2s 6ms/step - loss: 2.4709 - val_loss: 0.4371\n",
      "Epoch 23/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3770 - val_loss: 0.4150\n",
      "Epoch 24/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3747 - val_loss: 0.4334\n",
      "Epoch 25/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3649 - val_loss: 0.4418\n",
      "Epoch 26/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.4111 - val_loss: 0.5194\n",
      "Epoch 27/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.4714 - val_loss: 0.4980\n",
      "Epoch 28/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.4562 - val_loss: 0.5366\n",
      "Epoch 29/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.4180 - val_loss: 0.6887\n",
      "Epoch 30/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.7085 - val_loss: 0.4709\n",
      "Epoch 31/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3461 - val_loss: 0.3885\n",
      "Epoch 32/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3513 - val_loss: 0.4275\n",
      "Epoch 33/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3690 - val_loss: 0.3841\n",
      "Epoch 34/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3606 - val_loss: 0.4279\n",
      "Epoch 35/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3853 - val_loss: 0.3614\n",
      "Epoch 36/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3352 - val_loss: 0.4280\n",
      "Epoch 37/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3472 - val_loss: 0.4062\n",
      "Epoch 38/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3630 - val_loss: 0.3840\n",
      "Epoch 39/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3172 - val_loss: 0.3687\n",
      "Epoch 40/40\n",
      "363/363 [==============================] - 2s 5ms/step - loss: 0.3310 - val_loss: 0.3562\n",
      "162/162 [==============================] - 0s 2ms/step - loss: 0.3119\n"
     ]
    }
   ],
   "source": [
    "model = ResidualRegressor(1)\n",
    "early_stopping_cb = keras.callbacks.EarlyStopping(patience= 10, restore_best_weights=True)\n",
    "model.compile(loss=\"mse\", optimizer=\"nadam\")\n",
    "history = model.fit(X_train_scaled, y_train, epochs=40, validation_data=(X_valid_scaled, y_valid),callbacks=[early_stopping_cb])\n",
    "score = model.evaluate(X_test_scaled, y_test)\n",
    "y_pred = model.predict(X_test_scaled)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.6.9 64-bit ('venv': venv)",
   "language": "python",
   "name": "python36964bitvenvvenva66490d71b884a119226d82deedf3358"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}