14_02_multilayer-networks.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## How a regression network is traditionally trained\n",
    "\n",
    "This network is trained using a data set $D = ({{\\bf x}^{(n)}, {\\bf t}^{(n)}})$ by adjusting ${\\bf w}$ so as to minimize an error function, e.g.,\n",
    "\n",
    "$$\n",
    "E_D({\\bf w}) = \\sum_n\\sum_i (y_i({\\bf x}^{(n)};{\\bf w}) - t_i^{(n)})^2\n",
    "$$\n",
    "\n",
    "This objective function is a sum of terms, one for each input/target pair $\\{ {\\bf x}, {\\bf t} \\}$, measuring how close the output ${\\bf y}({\\bf x}; {\\bf w})$ is to the target ${\\bf t}$:\n",
    "\n",
    "$$\n",
    "E_D({\\bf w}) = \\sum_n E_{\\bf x}^{(n)}, \\quad E_{\\bf x}^{(n)}=\\sum_i (y_i({\\bf x}^{(n)};{\\bf w}) - t_i^{(n)})^2\n",
    "$$\n",
    "\n",
    "\n",
    "\n",
    "This minimization is based on repeated evaluation of the gradient of $E_D$. This gradient can be efficiently computed using the backpropagation algorithm which uses the chain rule to find the derivatives, as we discuss below.\n",
    "\n",
    "Often, regularization (also known as weight decay) is included, modifying\n",
    "the objective function to:\n",
    "\n",
    "$$\n",
    "M({\\bf w})=\\alpha E_D({\\bf w}) + \\beta E_W({\\bf w}),\n",
    "$$\n",
    "where $E_W = \\frac{1}{2}\\sum_i w_i^2$.\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "### Gradient descent\n",
    "(From Wikipedia)\n",
    "Cool animations at http://www.benfrederickson.com/numerical-optimization/\n",
    "\n",
    "Gradient descent is a first-order iterative optimization algorithm for finding the minimum of a function. To find a local minimum of a function using gradient descent, one takes steps proportional to the negative of the gradient (or of the approximate gradient) of the function at the current point.\n",
    "\n",
    "Gradient descent is based on the observation that if the multi-variable function $ F(\\mathbf {x} )$ is defined and differentiable in a neighborhood of a point $ \\mathbf {a}$ , then $ F(\\mathbf {x} )$ decreases fastest if one goes from $ \\mathbf {a}$ in the direction of the negative gradient of $F$ at $ \\mathbf {a}$ , $ -\\nabla F(\\mathbf {a} )$. It follows that, if\n",
    "\n",
    "$$\\mathbf {a} _{n+1}=\\mathbf {a} _{n}-\\eta \\nabla F(\\mathbf {a} _{n})$$ \n",
    "\n",
    "for $\\eta$  small enough, then $F(\\mathbf {a_{n}} )\\geq F(\\mathbf {a_{n+1}} )$. In other words, the term $\\eta \\nabla F(\\mathbf {a} )$ is subtracted from $ \\mathbf {a}$  because we want to move against the gradient, namely down toward the minimum. With this observation in mind, one starts with a guess $\\mathbf {x} _{0}$ for a local minimum of $F$, and considers the sequence $\\mathbf {x} _{0},\\mathbf {x} _{1},\\mathbf {x} _{2},\\dots$  such that\n",
    "\n",
    "$${x} _{n+1}=\\mathbf {x} _{n}-\\gamma _{n}\\nabla F(\\mathbf {x} _{n}),\\ n\\geq 0.$$\n",
    "\n",
    "We have\n",
    "\n",
    "$F(\\mathbf {x} _{0})\\geq F(\\mathbf {x} _{1})\\geq F(\\mathbf {x} _{2})\\geq \\cdots$ ,\n",
    "so hopefully the sequence $(\\mathbf {x} _{n})$ converges to the desired local minimum. Note that the value of the step size $\\eta$  is allowed to change at every iteration.\n",
    "\n",
    "This process is illustrated in the adjacent picture. Here $F$ is assumed to be defined on the plane, and that its graph has a bowl shape. The blue curves are the contour lines, that is, the regions on which the value of $F$ is constant. A red arrow originating at a point shows the direction of the negative gradient at that point. Note that the (negative) gradient at a point is orthogonal to the contour line going through that point. We see that gradient descent leads us to the bottom of the bowl, that is, to the point where the value of the function $F$ is minimal.\n",
    "\n",
    "<img src=\"figures/Gradient_descent.png\" style=\"width: 350px;\"/>\n",
    "#### Illustration of the gradient descept procedure on a series of iterations down a bowl shaped surface\n",
    "\n",
    "The \"Zig-Zagging\" nature of the method is also evident below, where the gradient descent method is applied to $$F(x,y)=\\sin \\left({\\frac {1}{2}}x^{2}-{\\frac {1}{4}}y^{2}+3\\right)\\cos(2x+1-e^{y})$$\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.3226478037930326 1.602369170618785\n"
     ]
    },
    {
     "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": [
    "%matplotlib inline\n",
    "from matplotlib import pyplot\n",
    "pyplot.rcParams['image.cmap'] = 'jet'\n",
    "import numpy as np\n",
    "\n",
    "x0 = -1.4\n",
    "y0 = 0.5\n",
    "x = [x0] # The algorithm starts at x0, y0\n",
    "y = [y0] \n",
    "\n",
    "eta = 0.1 # step size multiplier\n",
    "precision = 0.00001\n",
    "\n",
    "def f(x,y):\n",
    "    f1 = x**2/2-y**2/4+3\n",
    "    f2 = 2*x+1-np.exp(y)\n",
    "    return np.sin(f1)*np.cos(f2)\n",
    "\n",
    "def gradf(x,y):\n",
    "    f1 = x**2/2-y**2/4+3\n",
    "    f2 = 2*x+1-np.exp(y)\n",
    "    dx = np.cos(f1)*np.cos(f2)*x-np.sin(f1)*np.sin(f2)*2.\n",
    "    dy = np.cos(f1)*np.cos(f2)*(-y/2.)-np.sin(f1)*np.sin(f2)*(-np.exp(y))\n",
    "    return (dx,dy)\n",
    "\n",
    "err = 100.\n",
    "while err > precision:\n",
    "    (step_x, step_y) = gradf(x0, y0)\n",
    "    x0 -= eta*step_x\n",
    "    y0 -= eta*step_y\n",
    "    x.append(x0)\n",
    "    y.append(y0)\n",
    "    err = eta*(abs(step_x)+abs(step_y))\n",
    "\n",
    "\n",
    "print(x0,y0)\n",
    "\n",
    "#### All this below is just to visualize the process\n",
    "dx = 0.05\n",
    "dy = 0.05\n",
    "xx = np.arange(-1.5, 1.+dx, dx)\n",
    "yy = np.arange(0., 2.+dy, dy)\n",
    "V = np.zeros(shape=(len(yy),len(xx)))\n",
    "\n",
    "for iy in range(0,len(yy)):\n",
    "    for ix in range(0,len(xx)):\n",
    "        V[iy,ix] = f(xx[ix],yy[iy])\n",
    "\n",
    "X, Y = np.meshgrid(xx, yy)\n",
    "pyplot.contour(X, Y, V)\n",
    "\n",
    "#pyplot.plot(x,y,linestyle='--', lw=3);\n",
    "pyplot.scatter(x,y);\n",
    "\n",
    "pyplot.ylabel(\"y\")\n",
    "pyplot.xlabel(\"x\");"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Stochastic gradient descent (SGD)\n",
    "\n",
    "Stochastic gradient descent (often shortened to SGD), also known as incremental gradient descent, is a stochastic approximation of the gradient descent optimization and iterative method for minimizing an objective function that is written as a sum of differentiable functions. \n",
    "\n",
    "There are a number of challenges in applying the gradient descent rule. To understand what the problem is, let's look back at the quadratic cost $E_D$. Notice that this cost function has the form $E=\\sum_n E_{\\bf x}^{(n)}$\n",
    "In practice, to compute the gradient $\\nabla E_D$\n",
    " we need to compute the gradients $\\nabla E_{\\bf x}^{(n)}$\n",
    " separately for each training input, ${\\bf x^{(n)}}$\n",
    "and then average them.\n",
    ". Unfortunately, when the number of training inputs is very large this can take a long time, and learning thus occurs slowly.\n",
    "\n",
    "Stochastic gradient descent can be used to speed up learning. The idea is to estimate the gradient $\\nabla E$\n",
    " by computing $\\nabla E_{\\bf x}$\n",
    "for a small sample of randomly chosen training inputs. By averaging over this small sample it turns out that we can quickly get a good estimate of the true gradient.\n",
    "\n",
    "<!--To make these ideas more precise, stochastic gradient descent works by randomly picking out a small number $m$\n",
    " of randomly chosen training inputs. We'll label those random training inputs ${\\bf x^{(1)},x^{(2)},…,x^{(m)}}$\n",
    ", and refer to them as a mini-batch. Provided the sample size m\n",
    " is large enough we expect that the average value of the $\\nabla E_x$\n",
    " will be roughly equal to the average over all of them, that is\n",
    "$$\\frac{1}{m}\\sum _{j=1}^m \\nabla E_{x^{j}} \\approx \\frac{1}{n}\\sum _{j=1}^n \\nabla E_{x^{j}}$$\n",
    "where the second sum is over the entire set of training data. \n",
    "!-->\n",
    "\n",
    "To connect this explicitly to learning in neural networks, suppose $w_k$\n",
    " and $b_l$\n",
    " denote the weights and biases in our neural network. Then stochastic gradient descent works by picking out a randomly chosen mini-batch of training inputs, and training with those,\n",
    "$$\n",
    "w_k \\rightarrow w_k - \\eta \\sum_{j=1}^m \\frac{\\partial{E_{\\bf x}^{(j)}}}{\\partial w_k}\n",
    "$$\n",
    "\n",
    "$$\n",
    "b_l \\rightarrow b_l - \\eta \\sum_{j=1}^m \\frac{\\partial{E_{\\bf x}^{(j)}}}{\\partial b_l}\n",
    "$$\n",
    "\n",
    "where the sums are over all the training examples in the current mini-batch. Then we pick out another randomly chosen mini-batch and train with those. And so on, until we have exhausted the training inputs, which is said to complete an epoch of training. At that point we start over with a new training epoch.\n",
    "\n",
    "The pseudocode would look like:\n",
    "\n",
    "`Choose an initial vector of parameters` $w$ `and learning rate` $\\eta$.\n",
    "\n",
    "`Repeat until an approximate minimum is obtained:`\n",
    "```\n",
    "    Randomly shuffle examples in the training set.\n",
    "    For i=1,2,...,n , do:\n",
    "```\n",
    "$\\quad \\quad \\quad \\quad \\quad w:=w-\\eta \\nabla E_{i}(w).$ \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Example: linear regression\n",
    "\n",
    "As seen previously, the objective function to be minimized is:\n",
    "\n",
    "$$\n",
    "\\begin{aligned}\n",
    "E(w)=\\sum _{i=1}^{n}E_{i}(w)=\\sum _{i=1}^{n}\\left(w_{1}+w_{2}x_{i}-y_{i}\\right)^{2}.\n",
    "\\end{aligned}\n",
    "$$\n",
    "\n",
    "And the gradent descent equations can be written in matrix form as:\n",
    "\n",
    "$$\n",
    "\\begin{bmatrix}w_{1}\\\\w_{2}\\end{bmatrix}:={\\begin{bmatrix}w_{1}\\\\w_{2}\\end{bmatrix}}-\\eta {\\begin{bmatrix}2(w_{1}+w_{2}x_{i}-y_{i})\\\\2x_{i}(w_{1}+w_{2}x_{i}-y_{i})\\end{bmatrix}}.\n",
    "$$\n",
    "\n",
    "We'll generate a series of 100 random points aligned more or less along the line $y=a+bx$ with $a=1$ and $b=2$\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.1637760980701564 2.001777141438794\n"
     ]
    },
    {
     "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": [
    "%matplotlib inline\n",
    "from matplotlib import pyplot\n",
    "import numpy as np\n",
    "\n",
    "a = 1\n",
    "b = 2\n",
    "num_points = 100\n",
    "np.random.seed(637163) # we make sure we always generate the same sequence\n",
    "x_data = np.random.rand(num_points)*20.\n",
    "y_data = x_data*b+a+3*(2.*np.random.rand(num_points)-1)\n",
    "\n",
    "pyplot.scatter(x_data,y_data)\n",
    "pyplot.plot(x_data, b*x_data+a)\n",
    "\n",
    "#### Least squares fit\n",
    "sum_x = np.sum(x_data)\n",
    "sum_y = np.sum(y_data)\n",
    "sum_x2 = np.sum(x_data**2)\n",
    "sum_xy = np.sum(x_data*y_data)\n",
    "det = num_points*sum_x2-sum_x**2\n",
    "fit_a = (sum_y*sum_x2-sum_x*sum_xy)/det\n",
    "fit_b = (num_points*sum_xy-sum_x*sum_y)/det\n",
    "print(fit_a,fit_b)\n",
    "\n",
    "pyplot.xlim(-1,22)\n",
    "pyplot.ylim(-1,24)\n",
    "pyplot.plot(x_data, fit_b*x_data+fit_a);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now write an SGD code for this problem. The training_data is a list of tuples `(x, y)` representing the training inputs and corresponding desired outputs. The variables `epochs` and `mini_batch_size` are what you'd expect - the number of epochs to train for, and the size of the mini-batches to use when sampling. `eta` is the learning rate, $\\eta$. If the optional argument `test_data` is supplied, then the program will evaluate the network after each epoch of training, and print out partial progress. This is useful for tracking progress, but slows things down substantially.\n",
    "\n",
    "The code works as follows. In each epoch, it starts by randomly shuffling the training data, and then partitions it into mini-batches of the appropriate size. This is an easy way of sampling randomly from the training data. Then for each `mini_batch` we apply a single step of gradient descent. This is done by the code `self.update_mini_batch(mini_batch, eta)`, which updates the coefficients according to a single iteration of gradient descent, using just the training data in `mini_batch`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0: 2.851091506514999 1.7876531781856742\n",
      "Epoch 1: 2.827178748563592 1.963573397229449\n",
      "Epoch 2: 2.7801714492150196 1.8619634110879069\n",
      "Epoch 3: 2.756668947213542 2.0288955459781213\n",
      "Epoch 4: 2.7076355088367525 1.880872788736252\n",
      "Epoch 5: 2.6709938639494206 1.9094861291196814\n",
      "Epoch 6: 2.6459565191117687 2.083968448182042\n",
      "Epoch 7: 2.592571941252409 1.867303977943171\n",
      "Epoch 8: 2.5668650012203917 1.9282486257504263\n",
      "Epoch 9: 2.5260855862259803 1.8254093290530535\n",
      "Epoch 10: 2.4977119308055653 1.8393082159024072\n",
      "Epoch 11: 2.461504007375818 1.8187446556619598\n",
      "Epoch 12: 2.440510503849123 1.9639665902712513\n",
      "Epoch 13: 2.4192661745417667 2.056408083610859\n",
      "Epoch 14: 2.3778520184589746 1.9200785689241224\n",
      "Epoch 15: 2.352636754920605 1.9557702061187\n",
      "Epoch 16: 2.32668522997869 1.985365726392641\n",
      "Epoch 17: 2.29093524056995 1.9302755099169442\n",
      "Epoch 18: 2.264373631096743 1.9014850942257417\n",
      "Epoch 19: 2.240116842945883 1.935758005663645\n",
      "Epoch 20: 2.2192529930462856 2.0084713102139746\n",
      "Epoch 21: 2.1882155215215175 1.954607851080249\n",
      "Epoch 22: 2.1610945265017794 1.871860736902566\n",
      "Epoch 23: 2.1460869019987685 2.0191055043901693\n",
      "Epoch 24: 2.1134743012209274 1.917120770897182\n",
      "Epoch 25: 2.093876809726992 1.9695147135445203\n",
      "Epoch 26: 2.0732373408257976 2.009659333749805\n",
      "Epoch 27: 2.0545056929658063 1.9990817647672674\n",
      "Epoch 28: 2.0358849885566888 2.0684880037349314\n",
      "Epoch 29: 2.0127961055519528 1.981549450254667\n",
      "Epoch 30: 1.9912110003344137 1.9953084240830377\n",
      "Epoch 31: 1.9702356965968113 1.9337881228020706\n",
      "Epoch 32: 1.958931435902377 1.9703136538892496\n",
      "Epoch 33: 1.9320635446309764 1.8826305504717291\n",
      "Epoch 34: 1.9183934975941435 1.9108781471703922\n",
      "Epoch 35: 1.9002148717519194 1.933657775282379\n",
      "Epoch 36: 1.8765188176622247 1.8501242260953568\n",
      "Epoch 37: 1.8636240317182582 1.8732118340436419\n",
      "Epoch 38: 1.8580542540860803 2.043103451641688\n",
      "Epoch 39: 1.8404926684574494 1.9833198887817372\n",
      "Epoch 40: 1.818471335258018 1.885853165755979\n",
      "Epoch 41: 1.8104060287044144 1.946463310907274\n",
      "Epoch 42: 1.7945224315136403 1.955383791204891\n",
      "Epoch 43: 1.782283463654412 2.0023405102821905\n",
      "Epoch 44: 1.749520717458214 1.8281773503780525\n",
      "Epoch 45: 1.7453813582714632 2.003640838074842\n",
      "Epoch 46: 1.728623006260412 1.9766876117635683\n",
      "Epoch 47: 1.7084093749222484 1.934436123013543\n",
      "Epoch 48: 1.6882385559184097 1.8316669036210635\n",
      "Epoch 49: 1.6872347550750442 1.9027956243651138\n",
      "Epoch 50: 1.6799350406816318 2.0067968405782595\n",
      "Epoch 51: 1.6544625765529073 1.8881985255370595\n",
      "Epoch 52: 1.6437316979583543 1.9255046345948543\n",
      "Epoch 53: 1.6381622976293797 2.0030543868483233\n",
      "Epoch 54: 1.619288123392432 1.9054952135108394\n",
      "Epoch 55: 1.6081107386909623 1.8842249840685168\n",
      "Epoch 56: 1.606385744453914 1.995354114039655\n",
      "Epoch 57: 1.590951010569201 1.935539287963154\n",
      "Epoch 58: 1.5907159977092078 2.0039496119285936\n",
      "Epoch 59: 1.583681348958318 2.0712979889802097\n",
      "Epoch 60: 1.5630265005446518 1.9298084540897775\n",
      "Epoch 61: 1.564720140787769 2.0808310251200517\n",
      "Epoch 62: 1.5466686232306155 1.9187231079896427\n",
      "Epoch 63: 1.5388524762367308 1.9621349285397989\n",
      "Epoch 64: 1.5405490598559062 2.0291358831335318\n",
      "Epoch 65: 1.5208453821228052 1.8248931026367041\n",
      "Epoch 66: 1.5213798489903287 1.8930706301200002\n",
      "Epoch 67: 1.529153914875216 2.1490931535209667\n",
      "Epoch 68: 1.5022716493694421 1.9262311478789624\n",
      "Epoch 69: 1.500068778772673 2.0144696446422423\n",
      "Epoch 70: 1.4895292591761131 1.9988696565984023\n",
      "Epoch 71: 1.4750010294727158 1.9334703018634691\n",
      "Epoch 72: 1.468734739081002 1.9509592360886463\n",
      "Epoch 73: 1.4602541491491496 1.9535588676349709\n",
      "Epoch 74: 1.4629985193934691 2.087981232637853\n",
      "Epoch 75: 1.4482903637054756 1.9769627087994208\n",
      "Epoch 76: 1.4441169372446487 1.9911884358343\n",
      "Epoch 77: 1.4374914169928699 1.9697513594833118\n",
      "Epoch 78: 1.4290881203003065 1.9810815298729503\n",
      "Epoch 79: 1.4246361485906045 2.021522320675344\n",
      "Epoch 80: 1.416276663539712 1.997969411686069\n",
      "Epoch 81: 1.4088098511794314 1.9930406863592567\n",
      "Epoch 82: 1.4068185837259672 2.0164782110987383\n",
      "Epoch 83: 1.3962324622369873 1.8977840351410191\n",
      "Epoch 84: 1.401436092664027 1.9950621893919915\n",
      "Epoch 85: 1.387471145834617 1.8824887521430227\n",
      "Epoch 86: 1.4014093218694679 2.1053832817280766\n",
      "Epoch 87: 1.3846125002362668 1.961879809671709\n",
      "Epoch 88: 1.3803176460734246 1.9693072109387666\n",
      "Epoch 89: 1.3733167787595022 1.9193370938253795\n",
      "Epoch 90: 1.3719395040783267 1.9529063749285855\n",
      "Epoch 91: 1.3657851409032489 1.9519469364018414\n",
      "Epoch 92: 1.365731695836303 2.0299060306472323\n",
      "Epoch 93: 1.352704037805701 1.9068562249348497\n",
      "Epoch 94: 1.358872635570531 1.9505330043142486\n",
      "Epoch 95: 1.352406342772186 1.924937800945274\n",
      "Epoch 96: 1.3529921471002395 1.9582904963736072\n",
      "Epoch 97: 1.3467058766687356 1.964421345479529\n",
      "Epoch 98: 1.3441056012495736 1.9585772301210544\n",
      "Epoch 99: 1.338307075761737 1.9641242330382802\n",
      "Epoch 100: 1.3389750343328595 2.0331089097262622\n",
      "Epoch 101: 1.3292056062449498 1.970316569830166\n",
      "Epoch 102: 1.3286221627701793 2.0548925899273556\n",
      "Epoch 103: 1.3249933581558049 2.0102278112773337\n",
      "Epoch 104: 1.3141363246433797 1.9167494489192494\n",
      "Epoch 105: 1.3112904405797088 1.9230166472931127\n",
      "Epoch 106: 1.3155463891174177 2.0363423473576034\n",
      "Epoch 107: 1.3066507352411303 2.002921226309002\n",
      "Epoch 108: 1.3012267385619987 1.992156333232618\n",
      "Epoch 109: 1.2998811762793303 2.0836475234636054\n",
      "Epoch 110: 1.2869768488136015 2.0326476069467847\n",
      "Epoch 111: 1.2782409042449163 1.955308271763259\n",
      "Epoch 112: 1.2702227407769626 1.843641680619616\n",
      "Epoch 113: 1.2841364346903865 2.0428258403939537\n",
      "Epoch 114: 1.2799771181355302 2.027691157552291\n",
      "Epoch 115: 1.272749198706809 1.9838436304512466\n",
      "Epoch 116: 1.2751164588056392 2.048106822415394\n",
      "Epoch 117: 1.2709851254712097 1.9917962774665723\n",
      "Epoch 118: 1.2762427149034599 2.136723976667836\n",
      "Epoch 119: 1.2614286952500717 1.9537916421283819\n",
      "Epoch 120: 1.268051811501334 2.003348786243271\n",
      "Epoch 121: 1.2705624559612398 2.05644114955335\n",
      "Epoch 122: 1.2488428579582682 1.824502349600589\n",
      "Epoch 123: 1.2609746890519609 2.046167589066578\n",
      "Epoch 124: 1.2649024819516497 2.073455837171682\n",
      "Epoch 125: 1.2553435081447994 1.9933812628787015\n",
      "Epoch 126: 1.2598887147760574 2.0313796669882813\n",
      "Epoch 127: 1.2599147404360962 2.062627907835911\n",
      "Epoch 128: 1.249692892431034 1.975233671612488\n",
      "Epoch 129: 1.2442021578463776 1.9552628960704093\n",
      "Epoch 130: 1.2386105150995876 1.982099666924853\n",
      "Epoch 131: 1.2355586187412317 1.9757114294223843\n",
      "Epoch 132: 1.2443294895643744 2.101623933536461\n",
      "Epoch 133: 1.2299004597831222 1.9768187344562476\n",
      "Epoch 134: 1.2276038542035699 1.9822626470879492\n",
      "Epoch 135: 1.2277446415672508 1.9615810105868339\n",
      "Epoch 136: 1.2286379676808046 2.0355041539229313\n",
      "Epoch 137: 1.2211714193515004 1.9701469343010707\n",
      "Epoch 138: 1.2298287656838378 2.024296033663489\n",
      "Epoch 139: 1.227733776450261 2.035072556742687\n",
      "Epoch 140: 1.2188039214924324 1.9586362914585522\n",
      "Epoch 141: 1.221958994511311 2.018941730208499\n",
      "Epoch 142: 1.2157233735073354 1.9671959386100284\n",
      "Epoch 143: 1.2124366156263169 1.9592948983993996\n",
      "Epoch 144: 1.2147369220284092 2.0442536562207776\n",
      "Epoch 145: 1.205955712341693 1.9694423842214128\n",
      "Epoch 146: 1.2049971023807025 1.9859694713761042\n",
      "Epoch 147: 1.2075221330905666 2.009002905561291\n",
      "Epoch 148: 1.2115221193191605 2.059401540299177\n",
      "Epoch 149: 1.2025709919887162 1.982096858804897\n",
      "Epoch 150: 1.2025820405123533 1.9921201608490595\n",
      "Epoch 151: 1.2044366839866598 2.0146807988967903\n",
      "Epoch 152: 1.206892541459285 2.037165558406217\n",
      "Epoch 153: 1.2084186753224047 2.0514455070803828\n",
      "Epoch 154: 1.2089888077827122 2.051536688143954\n",
      "Epoch 155: 1.2076227373011457 1.9988099365677854\n",
      "Epoch 156: 1.2037083483319981 1.9950064987376799\n",
      "Epoch 157: 1.206526240140426 2.074083444964129\n",
      "Epoch 158: 1.1969956856897452 1.9783641111066796\n",
      "Epoch 159: 1.1988308898232631 1.9543139803979717\n",
      "Epoch 160: 1.1995691446186438 1.937559081054288\n",
      "Epoch 161: 1.2038438870466412 2.015030834060468\n",
      "Epoch 162: 1.1985588246541923 1.9698469983800948\n",
      "Epoch 163: 1.1958191873606725 1.9663497539993224\n",
      "Epoch 164: 1.186928424021493 1.86455626410724\n",
      "Epoch 165: 1.1928061814986093 2.0497622851939536\n",
      "Epoch 166: 1.1865071366714997 2.0408127489658647\n",
      "Epoch 167: 1.182851618741269 2.0110138481333046\n",
      "Epoch 168: 1.18782422094513 2.0552683461751196\n",
      "Epoch 169: 1.1897908718277397 2.0792826146138967\n",
      "Epoch 170: 1.1895586261808533 2.020543573678281\n",
      "Epoch 171: 1.1825696897860893 1.9801100085432106\n",
      "Epoch 172: 1.181820009229163 1.9799370528048312\n",
      "Epoch 173: 1.1868381738615787 2.0495345787064054\n",
      "Epoch 174: 1.1839766393778706 2.0473350252553804\n",
      "Epoch 175: 1.1850485909968755 2.0716526425398674\n",
      "Epoch 176: 1.1824535090862134 2.011224705734455\n",
      "Epoch 177: 1.180767591043016 2.004658322219915\n",
      "Epoch 178: 1.1821388960311063 1.988800626421598\n",
      "Epoch 179: 1.1880522464685088 2.067303807558048\n",
      "Epoch 180: 1.1759645896758149 1.9546667268531908\n",
      "Epoch 181: 1.1790533790141011 1.9842002311282236\n",
      "Epoch 182: 1.1836537100256175 2.0460500757626106\n",
      "Epoch 183: 1.1822863317308494 2.025085525889144\n",
      "Epoch 184: 1.172530841497514 1.9242741669889663\n",
      "Epoch 185: 1.1823675397686417 2.0441537662254428\n",
      "Epoch 186: 1.177468047390862 1.9588840402287566\n",
      "Epoch 187: 1.1801243667511367 1.970880411546426\n",
      "Epoch 188: 1.1867122725568682 2.071167968819989\n",
      "Epoch 189: 1.175747819017282 1.9120770900835262\n",
      "Epoch 190: 1.178488236180694 1.9711391056919325\n",
      "Epoch 191: 1.1834159610241293 2.076521736901644\n",
      "Epoch 192: 1.1840720138080807 2.0695169723793168\n",
      "Epoch 193: 1.1878662150844443 2.1054361965492254\n",
      "Epoch 194: 1.172204080715583 1.9193349078207402\n",
      "Epoch 195: 1.1796266578616825 2.021273557852105\n",
      "Epoch 196: 1.1756283433183137 1.978397476038657\n",
      "Epoch 197: 1.179347870801012 2.057485247243262\n",
      "Epoch 198: 1.1771510299609378 1.9839566352038045\n",
      "Epoch 199: 1.1734142144784492 1.9413638781685132\n",
      "Epoch 200: 1.1796858491714215 2.0257042728471295\n",
      "Epoch 201: 1.1828027565096388 2.073049797618982\n",
      "Epoch 202: 1.1836229294305753 2.105296552012404\n",
      "Epoch 203: 1.1748330785870187 1.97842429934267\n",
      "Epoch 204: 1.1710094849765988 1.9225888313967678\n",
      "Epoch 205: 1.1750019318878657 2.0161089101204266\n",
      "Epoch 206: 1.1688822960668843 1.9173751852467291\n",
      "Epoch 207: 1.171000593729568 1.9496922473163611\n",
      "Epoch 208: 1.1719395299814825 1.9846125819542637\n",
      "Epoch 209: 1.1653718472989494 1.9657401115851765\n",
      "Epoch 210: 1.1666777708673137 1.9969331887938948\n",
      "Epoch 211: 1.1647159687365278 1.9909138128673325\n",
      "Epoch 212: 1.1754452887943576 2.0857264475141384\n",
      "Epoch 213: 1.169688967309767 2.054910054131244\n",
      "Epoch 214: 1.169127580623336 2.0601206215450856\n",
      "Epoch 215: 1.1652074626875797 1.9973034692573273\n",
      "Epoch 216: 1.1702852193446573 2.0519593792424358\n",
      "Epoch 217: 1.1646162669879867 1.968465416447997\n",
      "Epoch 218: 1.1707704669426298 2.0249056512171877\n",
      "Epoch 219: 1.1724593556038376 2.082380488563087\n",
      "Epoch 220: 1.1600188422806863 1.9533301129439185\n",
      "Epoch 221: 1.1603990597387488 1.9908556837124307\n",
      "Epoch 222: 1.172306977616543 2.149832223460105\n",
      "Epoch 223: 1.1533139870665958 1.897368373790175\n",
      "Epoch 224: 1.1603077278460978 1.9740605449537711\n",
      "Epoch 225: 1.1610865668922927 2.0156478512874445\n",
      "Epoch 226: 1.1598149090178131 2.0062030866511695\n",
      "Epoch 227: 1.1629546106433142 2.010542546698093\n",
      "Epoch 228: 1.1636366920994934 2.0645587753976815\n",
      "Epoch 229: 1.158441789962946 1.9732275213983936\n",
      "Epoch 230: 1.1594068941620521 2.0211588360306543\n",
      "Epoch 231: 1.157992091904049 1.9899786212760588\n",
      "Epoch 232: 1.1586025563761901 1.9669302651828673\n",
      "Epoch 233: 1.158935020774732 1.9624153218725968\n",
      "Epoch 234: 1.1780146089004881 2.2084053499839253\n",
      "Epoch 235: 1.1624086023948017 1.9904183594049172\n",
      "Epoch 236: 1.1687114571640016 2.0243325338089573\n",
      "Epoch 237: 1.163780074230273 2.0115919918004996\n",
      "Epoch 238: 1.1593855584185195 1.9610645305963619\n",
      "Epoch 239: 1.1557913057904843 1.9313205589116578\n",
      "Epoch 240: 1.1604861873712111 2.0147368305307953\n",
      "Epoch 241: 1.1643855379493093 2.0308919296252905\n",
      "Epoch 242: 1.1684656530568671 2.0836439974908796\n",
      "Epoch 243: 1.156170351707462 1.9437864267698877\n",
      "Epoch 244: 1.1632490253614063 2.0331559937553387\n",
      "Epoch 245: 1.1625416949346654 1.9878681854250837\n",
      "Epoch 246: 1.1743225954897536 2.139668084828036\n",
      "Epoch 247: 1.160922425872473 1.9673081971292015\n",
      "Epoch 248: 1.158370283489093 1.9186952135315374\n",
      "Epoch 249: 1.1590132165288352 1.9575008370562035\n",
      "Epoch 250: 1.1692487934322693 2.047704473254694\n",
      "Epoch 251: 1.1661088708383454 2.0272174059642936\n",
      "Epoch 252: 1.164244940978814 1.9922728331857886\n",
      "Epoch 253: 1.1740138877833493 2.144034922963351\n",
      "Epoch 254: 1.1627478961724935 1.9765857849724182\n",
      "Epoch 255: 1.157342812461918 1.9176534632661586\n",
      "Epoch 256: 1.167310602022143 2.023937327632803\n",
      "Epoch 257: 1.1664092885855957 2.0225094450202428\n",
      "Epoch 258: 1.166403114048588 2.0214241796771435\n",
      "Epoch 259: 1.1640833336509528 1.9501414768095904\n",
      "Epoch 260: 1.172884748612123 2.047402314044348\n",
      "Epoch 261: 1.1733694938120844 2.0709007251204166\n",
      "Epoch 262: 1.1703883076248496 1.9846698128270954\n",
      "Epoch 263: 1.1721027556301264 2.037220774305596\n",
      "Epoch 264: 1.17839549649102 2.1112297457663614\n",
      "Epoch 265: 1.1654972084590798 1.9444411088542253\n",
      "Epoch 266: 1.1770102224348051 2.0984330461842684\n",
      "Epoch 267: 1.1661324837193872 1.9735288321449527\n",
      "Epoch 268: 1.1684228835987054 2.0136744531303976\n",
      "Epoch 269: 1.162849000284166 1.9527235425674798\n",
      "Epoch 270: 1.1734050786860557 2.0986326550647374\n",
      "Epoch 271: 1.160698086530356 1.968205613519547\n",
      "Epoch 272: 1.1718205074971704 2.087588337895705\n",
      "Epoch 273: 1.1707801461385712 2.05541134597421\n",
      "Epoch 274: 1.1691796980186806 2.0359205380058287\n",
      "Epoch 275: 1.1681172510742326 2.0465446260179836\n",
      "Epoch 276: 1.1668021632958008 2.055020724857271\n",
      "Epoch 277: 1.151646609766444 1.9027054110156603\n",
      "Epoch 278: 1.1630268716903074 2.0320960711046787\n",
      "Epoch 279: 1.1532204157848187 1.9267965144776447\n",
      "Epoch 280: 1.1545101829463142 1.9504711296639747\n",
      "Epoch 281: 1.1539752359172013 1.9155125002617734\n",
      "Epoch 282: 1.1605993185017476 1.9984865154380536\n",
      "Epoch 283: 1.16832157879913 2.0808959291191385\n",
      "Epoch 284: 1.1571226626221311 1.9496603321626576\n",
      "Epoch 285: 1.1637392503157475 1.966644596396405\n",
      "Epoch 286: 1.1703061954315055 2.0508837977273955\n",
      "Epoch 287: 1.1661373161151396 1.9581123052424907\n",
      "Epoch 288: 1.1716489545640907 2.0660943356351633\n",
      "Epoch 289: 1.162210911845686 2.000755633795154\n",
      "Epoch 290: 1.1665233458674846 2.103888995353846\n",
      "Epoch 291: 1.152381620806598 2.041867334236522\n",
      "Epoch 292: 1.1533313698555427 2.0270061674169306\n",
      "Epoch 293: 1.1540479557480243 2.0564307797064973\n",
      "Epoch 294: 1.1530786586022215 2.015652244367046\n",
      "Epoch 295: 1.1586921442683673 2.0977453777519637\n",
      "Epoch 296: 1.1466277814629358 1.9802845942481904\n",
      "Epoch 297: 1.1484481049977502 1.9713593809492327\n",
      "Epoch 298: 1.145805258441764 1.9447097646418092\n",
      "Epoch 299: 1.1476998763032493 1.9471699750330864\n",
      "Epoch 300: 1.1475673482226003 1.946809335925523\n",
      "Epoch 301: 1.1450321497622773 1.9366124455646785\n",
      "Epoch 302: 1.1610396227941684 2.084448184315205\n",
      "Epoch 303: 1.15617334776939 2.066528471403558\n",
      "Epoch 304: 1.157186834932992 2.039126665822382\n",
      "Epoch 305: 1.1518226564498195 2.0724202431858814\n",
      "Epoch 306: 1.1475072952738203 2.0187104774884315\n",
      "Epoch 307: 1.144911046485268 1.9691664109997982\n",
      "Epoch 308: 1.1436634433193602 1.9981925681435333\n",
      "Epoch 309: 1.1440556245610136 1.9958492184228103\n",
      "Epoch 310: 1.1416845744986281 1.9567346183934688\n",
      "Epoch 311: 1.1465242589209208 1.9965803402508104\n",
      "Epoch 312: 1.1402822568281583 1.9572394426584003\n",
      "Epoch 313: 1.1444256744712749 1.9917700404530183\n",
      "Epoch 314: 1.1341670575342746 1.8563959255071167\n",
      "Epoch 315: 1.1451735809680927 1.9990284579202906\n",
      "Epoch 316: 1.1448462684191392 1.946829763991842\n",
      "Epoch 317: 1.1460292976769055 1.9591527701272857\n",
      "Epoch 318: 1.1524783638351412 2.0395256491258884\n",
      "Epoch 319: 1.1481255014746177 1.955782751168601\n",
      "Epoch 320: 1.1535458673634562 2.018376299089275\n",
      "Epoch 321: 1.1564232403894963 2.077434874460244\n",
      "Epoch 322: 1.147254973345927 1.9583338917817799\n",
      "Epoch 323: 1.1516395714656882 2.009468785579987\n",
      "Epoch 324: 1.1557673220619185 2.063228215257118\n",
      "Epoch 325: 1.1534744287737841 1.9935416955109886\n",
      "Epoch 326: 1.150181727132005 1.949609916674194\n",
      "Epoch 327: 1.1441988979348574 1.869514673151032\n",
      "Epoch 328: 1.1501336543379006 1.9673135310095184\n",
      "Epoch 329: 1.1511906713198927 1.9199337926780946\n",
      "Epoch 330: 1.1494617877918734 1.9073749427857871\n",
      "Epoch 331: 1.1469113985657917 1.9805302170310206\n",
      "Epoch 332: 1.1541405113958176 2.0325597852045463\n",
      "Epoch 333: 1.1402980906326248 1.8390940750249538\n",
      "Epoch 334: 1.1531753874448596 1.9989140653485673\n",
      "Epoch 335: 1.1567599623902975 2.017653960628913\n",
      "Epoch 336: 1.1581622391493147 2.015007567430858\n",
      "Epoch 337: 1.1590143212810402 2.0609980775168237\n",
      "Epoch 338: 1.1510751732507323 2.001375491561636\n",
      "Epoch 339: 1.1481402813804618 1.9660002677298571\n",
      "Epoch 340: 1.1493973273100468 1.9400570063599067\n",
      "Epoch 341: 1.1583282367595962 2.0790651999854237\n",
      "Epoch 342: 1.1546969443141344 2.019734985064594\n",
      "Epoch 343: 1.1527831700666127 1.9674135205163883\n",
      "Epoch 344: 1.1549645414004304 2.0057042393205355\n",
      "Epoch 345: 1.1576926450047145 2.0576406101941864\n",
      "Epoch 346: 1.161981814267257 2.0685419449014533\n",
      "Epoch 347: 1.1618604580758003 2.0649695637689844\n",
      "Epoch 348: 1.1562319540905006 1.9578249376360854\n",
      "Epoch 349: 1.1625951733893936 1.9825105915855945\n",
      "Epoch 350: 1.157983377883249 1.9215557186068493\n",
      "Epoch 351: 1.1642282348431414 1.9658373090480363\n",
      "Epoch 352: 1.1528841361924658 1.8699415500500947\n",
      "Epoch 353: 1.1588619403644715 1.973303504547195\n",
      "Epoch 354: 1.1644781193218436 2.0546850735644964\n",
      "Epoch 355: 1.1603904375276721 2.025002971930877\n",
      "Epoch 356: 1.1549509698733604 1.9856711113014198\n",
      "Epoch 357: 1.1448840032833865 1.8643261327266034\n",
      "Epoch 358: 1.1526553868308187 1.9910940344948775\n",
      "Epoch 359: 1.1590624478940748 2.0905241266197585\n",
      "Epoch 360: 1.1438425028437778 1.8713452738982141\n",
      "Epoch 361: 1.1401137304292206 1.8118780505269045\n",
      "Epoch 362: 1.1495959026983391 1.9370362546355204\n",
      "Epoch 363: 1.1484417214953213 1.9652754975670976\n",
      "Epoch 364: 1.1488545201409164 1.990225168446491\n",
      "Epoch 365: 1.1446320914497148 1.9126347675717308\n",
      "Epoch 366: 1.149701200028218 1.9804858077778578\n",
      "Epoch 367: 1.143652647759062 1.867703685220218\n",
      "Epoch 368: 1.156057605455258 2.0361489292896566\n",
      "Epoch 369: 1.1523585704588584 2.0066252854364075\n",
      "Epoch 370: 1.1545886779431278 2.0608739349648895\n",
      "Epoch 371: 1.1513732490007436 1.9781191873402264\n",
      "Epoch 372: 1.153112913284324 1.9985838970000227\n",
      "Epoch 373: 1.1520276656465205 1.9937025191088742\n",
      "Epoch 374: 1.153933670869458 2.031596796668612\n",
      "Epoch 375: 1.1538351175449262 2.037474678249571\n",
      "Epoch 376: 1.1520646242862507 2.017049789453241\n",
      "Epoch 377: 1.1466829516420303 1.9419215391653144\n",
      "Epoch 378: 1.163010896069329 2.079237976454214\n",
      "Epoch 379: 1.1500419394047414 1.9335876359293516\n",
      "Epoch 380: 1.157359178919251 2.0173619548181474\n",
      "Epoch 381: 1.1553387148870204 1.9869264142006289\n",
      "Epoch 382: 1.150777670711536 1.89310014361363\n",
      "Epoch 383: 1.1604969603315158 2.0407271442227106\n",
      "Epoch 384: 1.1523358843246543 1.9426426565179926\n",
      "Epoch 385: 1.1635756840442424 2.004076157846764\n",
      "Epoch 386: 1.1673614420569172 2.0596411901504927\n",
      "Epoch 387: 1.1585544994201846 1.9538972375490153\n",
      "Epoch 388: 1.1635871495397658 2.02419123848167\n",
      "Epoch 389: 1.1626716170881763 1.9844917942397182\n",
      "Epoch 390: 1.1640985607676797 1.9942870242566482\n",
      "Epoch 391: 1.1667433246462153 2.012710781308925\n",
      "Epoch 392: 1.166057448474898 2.013157912048576\n",
      "Epoch 393: 1.1625924218958799 1.9780533169102879\n",
      "Epoch 394: 1.1677546394375398 2.0679503119367917\n",
      "Epoch 395: 1.165376031086863 2.040136965156445\n",
      "Epoch 396: 1.158296853822276 1.9564699927587872\n",
      "Epoch 397: 1.1606700470871392 1.9480817414503098\n",
      "Epoch 398: 1.168912601731494 2.041150736836026\n",
      "Epoch 399: 1.1706081338896337 2.021184077249987\n",
      "Epoch 400: 1.1801842754484062 2.1239415606354117\n",
      "Epoch 401: 1.1677212057163933 2.012395110885921\n",
      "Epoch 402: 1.1672601497312562 2.0311179189639055\n",
      "Epoch 403: 1.1446005340238736 1.745576950488077\n",
      "Epoch 404: 1.16644155750266 2.0371263086647105\n",
      "Epoch 405: 1.160096559777111 1.9573745798179463\n",
      "Epoch 406: 1.1561184001485691 1.9270445802835297\n",
      "Epoch 407: 1.1564428794892025 1.9534130611724356\n",
      "Epoch 408: 1.157832333324879 1.9990185814286516\n",
      "Epoch 409: 1.1542536578440337 1.9603170999631656\n",
      "Epoch 410: 1.1579641202641253 2.0197871817756243\n",
      "Epoch 411: 1.1483763991029288 1.9513250981321313\n",
      "Epoch 412: 1.1404922532824875 1.9209794035824443\n",
      "Epoch 413: 1.1507849734020454 2.0644281753554914\n",
      "Epoch 414: 1.1522729146307125 2.014082678488246\n",
      "Epoch 415: 1.1539226907918823 2.0452288186933716\n",
      "Epoch 416: 1.1553935535896283 2.0516110385635793\n",
      "Epoch 417: 1.1522612541246897 2.0132793132417066\n",
      "Epoch 418: 1.1445911719870643 1.8998976720242748\n",
      "Epoch 419: 1.15111755242973 1.9942442880638716\n",
      "Epoch 420: 1.1492918953099753 1.9723618584340097\n",
      "Epoch 421: 1.1508398588156288 2.0285906972912917\n",
      "Epoch 422: 1.1457613876618533 1.9442137226381164\n",
      "Epoch 423: 1.1544611539760208 2.033125521968535\n",
      "Epoch 424: 1.158597511625977 2.0883628018192923\n",
      "Epoch 425: 1.1571397943655337 2.030533907318707\n",
      "Epoch 426: 1.1485143165837068 1.9784233945860508\n",
      "Epoch 427: 1.154701391933309 1.9971076092250635\n",
      "Epoch 428: 1.153064423920393 1.9959524721314952\n",
      "Epoch 429: 1.1567851179708466 2.019159503275338\n",
      "Epoch 430: 1.1571300814502 2.0130031186865063\n",
      "Epoch 431: 1.1557711599737877 1.961595185581823\n",
      "Epoch 432: 1.1546534087018567 1.9417015113839\n",
      "Epoch 433: 1.1600875830711923 1.9985945740120918\n",
      "Epoch 434: 1.1621550174809294 2.0081173436065765\n",
      "Epoch 435: 1.1647858119071102 2.0069931180021467\n",
      "Epoch 436: 1.171922539030513 2.0754662230190997\n",
      "Epoch 437: 1.167478389331363 1.9709286580435443\n",
      "Epoch 438: 1.1584175000364914 1.8390082675323287\n",
      "Epoch 439: 1.1702472757144489 1.9891024656575065\n",
      "Epoch 440: 1.171445495576249 2.0087085929409083\n",
      "Epoch 441: 1.1707096902503504 1.9762975241864922\n",
      "Epoch 442: 1.1637849791651869 1.9026433988163767\n",
      "Epoch 443: 1.171386642660623 1.9761437663208812\n",
      "Epoch 444: 1.1779935021986057 2.046505549169825\n",
      "Epoch 445: 1.1671231314509358 1.9652553164672595\n",
      "Epoch 446: 1.1684116342751572 2.0495697045197634\n",
      "Epoch 447: 1.1647003781266703 1.9978788500721527\n",
      "Epoch 448: 1.1622405847121584 1.9532584123270598\n",
      "Epoch 449: 1.163145617594456 1.97476781288562\n",
      "Epoch 450: 1.1660072402971797 1.9742775233108107\n",
      "Epoch 451: 1.1754628552358457 2.0751210976336836\n",
      "Epoch 452: 1.1735655090109642 2.0469302285938347\n",
      "Epoch 453: 1.169287003886958 2.002149720380278\n",
      "Epoch 454: 1.1635890708472378 1.9611595279040226\n",
      "Epoch 455: 1.1715093753124384 2.069328916004224\n",
      "Epoch 456: 1.1680462317101417 2.0319407602859423\n",
      "Epoch 457: 1.172094998018456 2.0318137992892273\n",
      "Epoch 458: 1.160355532966944 1.8619506742616743\n",
      "Epoch 459: 1.173269399075153 2.027583132221297\n",
      "Epoch 460: 1.1684714341198927 1.963550735443599\n",
      "Epoch 461: 1.170713830400304 2.044528597373633\n",
      "Epoch 462: 1.1771672554199253 2.152225397121842\n",
      "Epoch 463: 1.165837688657203 2.0185833196372154\n",
      "Epoch 464: 1.1737217995420748 2.1373669706733183\n",
      "Epoch 465: 1.1578672638403995 1.9871013310791956\n",
      "Epoch 466: 1.155718232390874 1.9557080632755361\n",
      "Epoch 467: 1.1557765390859782 1.9816049106330598\n",
      "Epoch 468: 1.1618268652269887 2.070405227570197\n",
      "Epoch 469: 1.161491829599467 2.055096537614472\n",
      "Epoch 470: 1.1501254958475469 1.9661249599952544\n",
      "Epoch 471: 1.1572012124584645 2.0097660872796412\n",
      "Epoch 472: 1.1589604474043982 2.0302610947822126\n",
      "Epoch 473: 1.1570301975614823 2.023791417809254\n",
      "Epoch 474: 1.154208144563905 1.9919144916032592\n",
      "Epoch 475: 1.1519935009767621 1.9600780058834755\n",
      "Epoch 476: 1.1523555948256758 1.9765881895154538\n",
      "Epoch 477: 1.157402986756551 1.9904666517287626\n",
      "Epoch 478: 1.1633983269171395 2.052477744992646\n",
      "Epoch 479: 1.1599612891715394 1.977522581958186\n",
      "Epoch 480: 1.1630221140374917 2.0175865662949612\n",
      "Epoch 481: 1.1603236425247734 2.0134881202380304\n",
      "Epoch 482: 1.1622428147093367 2.041719852029051\n",
      "Epoch 483: 1.1663182676893415 2.053700356965903\n",
      "Epoch 484: 1.1715021454571897 2.1250912513664217\n",
      "Epoch 485: 1.1693220355043468 2.096288016694156\n",
      "Epoch 486: 1.162186265069271 2.007679351313094\n",
      "Epoch 487: 1.156178143974997 1.977949535688021\n",
      "Epoch 488: 1.1602071602457047 1.994003641212097\n",
      "Epoch 489: 1.1628305001357682 2.015400203164338\n",
      "Epoch 490: 1.161280006087401 2.003279628817588\n",
      "Epoch 491: 1.1599510493939122 2.002753704350214\n",
      "Epoch 492: 1.1601241418913693 1.98766468931923\n",
      "Epoch 493: 1.1622395894987398 1.9936463538182703\n",
      "Epoch 494: 1.157830575420715 1.9005282558070824\n",
      "Epoch 495: 1.1624944319141972 1.9714438683768225\n",
      "Epoch 496: 1.1622015660822151 1.9369276388742183\n",
      "Epoch 497: 1.1688861721589185 1.9886871017866947\n",
      "Epoch 498: 1.1696200308857707 2.0001774086437893\n",
      "Epoch 499: 1.1751592349446145 2.0805927706849436\n",
      "Epoch 500: 1.1726821241284315 2.0608937547750488\n",
      "Epoch 501: 1.1693244711262787 2.0588277961979427\n",
      "Epoch 502: 1.164318861403886 1.9568837861962476\n",
      "Epoch 503: 1.1674620721851714 2.001643097605654\n",
      "Epoch 504: 1.1686390240785633 2.0403778965299755\n",
      "Epoch 505: 1.160558346790356 1.9947000458052078\n",
      "Epoch 506: 1.1550101382922815 1.9553919771250783\n",
      "Epoch 507: 1.1633863462773795 2.047664305171632\n",
      "Epoch 508: 1.161234130215018 2.004647106635081\n",
      "Epoch 509: 1.163735588214571 2.0240105576567906\n",
      "Epoch 510: 1.154880290041301 1.9186242647350635\n",
      "Epoch 511: 1.165342490479345 2.0470743379697582\n",
      "Epoch 512: 1.1643114503808094 2.0418126189070493\n",
      "Epoch 513: 1.157790152976131 1.9940304144216299\n",
      "Epoch 514: 1.16612453996357 2.0405570445462846\n",
      "Epoch 515: 1.1624344094319572 1.9741268566492267\n",
      "Epoch 516: 1.1655260982797748 2.001969136200936\n",
      "Epoch 517: 1.1724045019844676 2.091767385080294\n",
      "Epoch 518: 1.1716510265738609 2.0843143345325106\n",
      "Epoch 519: 1.1686269961900118 2.073748595118983\n",
      "Epoch 520: 1.1638694305193875 2.0280956923017217\n",
      "Epoch 521: 1.1636480533268407 2.0332406500489677\n",
      "Epoch 522: 1.1603089324484057 1.9955956478220662\n",
      "Epoch 523: 1.1655246105364943 2.0737222183350674\n",
      "Epoch 524: 1.1662126905080017 2.076627691101107\n",
      "Epoch 525: 1.1568865515059628 1.974100652504382\n",
      "Epoch 526: 1.168518075087314 2.116093664338233\n",
      "Epoch 527: 1.1739213365871974 2.1025021700328166\n",
      "Epoch 528: 1.1640961527037137 2.0044410213216626\n",
      "Epoch 529: 1.1672718428903104 2.0129850523107686\n",
      "Epoch 530: 1.161172217703352 1.9262876678915872\n",
      "Epoch 531: 1.1585188388036571 1.9180818748253303\n",
      "Epoch 532: 1.1653540395516186 2.0225110646631754\n",
      "Epoch 533: 1.157019824202749 1.9078821673186914\n",
      "Epoch 534: 1.1661352704968753 2.005397500866252\n",
      "Epoch 535: 1.1627653941287703 1.9065647246932755\n",
      "Epoch 536: 1.1578514209429478 1.8395637428079765\n",
      "Epoch 537: 1.1662191899152756 2.00156943830715\n",
      "Epoch 538: 1.168201945668109 2.023471497362851\n",
      "Epoch 539: 1.1665228526276126 2.0644991500670145\n",
      "Epoch 540: 1.1658816613974488 2.058665702998422\n",
      "Epoch 541: 1.1643447355846515 1.986067069948219\n",
      "Epoch 542: 1.162476253404076 1.9906159550427986\n",
      "Epoch 543: 1.1747341048941513 2.15537441684528\n",
      "Epoch 544: 1.158303334439897 1.940222766571229\n",
      "Epoch 545: 1.167750097576072 2.047987879505625\n",
      "Epoch 546: 1.161499969447288 1.9796463160817586\n",
      "Epoch 547: 1.1618397809182983 2.0145491275750587\n",
      "Epoch 548: 1.1530248211180012 1.9542380628345206\n",
      "Epoch 549: 1.1531222017080704 1.9157917081991016\n",
      "Epoch 550: 1.1635858828490209 2.0209805841913973\n",
      "Epoch 551: 1.1579552887038964 1.9522790453630605\n",
      "Epoch 552: 1.1649831808558848 2.046686158613785\n",
      "Epoch 553: 1.1640074373615312 2.003159153474343\n",
      "Epoch 554: 1.1651457482691459 2.0102017022980583\n",
      "Epoch 555: 1.1715635029370517 2.0164581257660914\n",
      "Epoch 556: 1.1764307339401388 2.0724774274837436\n",
      "Epoch 557: 1.1669823043140815 1.9717705460185897\n",
      "Epoch 558: 1.1744484834107063 1.997186379174134\n",
      "Epoch 559: 1.169286604704071 1.9482003074049572\n",
      "Epoch 560: 1.1708232269684675 1.9670896160116247\n",
      "Epoch 561: 1.177918040159993 2.0833842383961665\n",
      "Epoch 562: 1.175450588919051 2.0409732812502694\n",
      "Epoch 563: 1.1757013259264324 2.036028007875955\n",
      "Epoch 564: 1.169852341967423 1.9178011529882681\n",
      "Epoch 565: 1.1839854524569744 2.1097700742101275\n",
      "Epoch 566: 1.1699901890362303 1.9443972610595948\n",
      "Epoch 567: 1.1784189477903166 2.0134108575047995\n",
      "Epoch 568: 1.1724111060404596 1.9501752190008588\n",
      "Epoch 569: 1.1793644855150411 2.058851187496867\n",
      "Epoch 570: 1.1710704634361726 1.9844881108378327\n",
      "Epoch 571: 1.170637349147944 2.0128491214203437\n",
      "Epoch 572: 1.1674859871552459 1.995827220735414\n",
      "Epoch 573: 1.1602092563525992 1.937681733563966\n",
      "Epoch 574: 1.1562973452235896 1.9081278689182501\n",
      "Epoch 575: 1.1601815573497287 1.9425865690783073\n",
      "Epoch 576: 1.1618195147540886 1.9472110697687859\n",
      "Epoch 577: 1.1708741364461133 2.043250021263253\n",
      "Epoch 578: 1.1676068093394385 2.003299759582118\n",
      "Epoch 579: 1.1651953384573908 1.9480119950750352\n",
      "Epoch 580: 1.1729481290413073 2.0578489946616196\n",
      "Epoch 581: 1.1756518105924687 2.081163220123037\n",
      "Epoch 582: 1.1665099587772048 1.9747648880773814\n",
      "Epoch 583: 1.167405836182548 1.9592179552107116\n",
      "Epoch 584: 1.1746998654284155 2.0028697591643874\n",
      "Epoch 585: 1.1773365248807184 2.0719121468963726\n",
      "Epoch 586: 1.174884035757471 2.0468189831104544\n",
      "Epoch 587: 1.166853113745297 1.9770294711982148\n",
      "Epoch 588: 1.164852928893286 1.9380153940723435\n",
      "Epoch 589: 1.1641376525904508 1.9377478395940595\n",
      "Epoch 590: 1.1599106893044446 1.9057579256872386\n",
      "Epoch 591: 1.1709976070406567 2.029604329245804\n",
      "Epoch 592: 1.1716528984848746 2.0380568684025455\n",
      "Epoch 593: 1.1686722556889133 2.0331959031481532\n",
      "Epoch 594: 1.1632339738919797 1.9820065033475909\n",
      "Epoch 595: 1.1564164096740799 1.9271097108640185\n",
      "Epoch 596: 1.1666694828568867 2.0724520045260295\n",
      "Epoch 597: 1.1537707827222201 1.9181065128036707\n",
      "Epoch 598: 1.1571075122251715 1.9705058791561187\n",
      "Epoch 599: 1.1721611168487915 2.1617614972119306\n",
      "Epoch 600: 1.155183953593385 1.9749704550186957\n",
      "Epoch 601: 1.1554366688928757 1.9844079087563904\n",
      "Epoch 602: 1.1647280640469377 2.081337702842052\n",
      "Epoch 603: 1.1658496108379754 2.1060936098291947\n",
      "Epoch 604: 1.14758503546801 1.8717290219119311\n",
      "Epoch 605: 1.1510322758027156 1.9657134666956575\n",
      "Epoch 606: 1.1535227944197683 1.989369661635029\n",
      "Epoch 607: 1.1538889452189671 2.0018450944565944\n",
      "Epoch 608: 1.1635852958354194 2.1179681261409202\n",
      "Epoch 609: 1.1554217599794048 1.9053692359601369\n",
      "Epoch 610: 1.1697281053661879 2.063683085157363\n",
      "Epoch 611: 1.1654273181045691 2.0749523365744253\n",
      "Epoch 612: 1.1532446422864804 1.9181586657893415\n",
      "Epoch 613: 1.163051758110536 2.046855736204119\n",
      "Epoch 614: 1.151106979641301 1.9490453567949784\n",
      "Epoch 615: 1.1589446331299682 2.0621825131187985\n",
      "Epoch 616: 1.156008845475296 2.0593705707766103\n",
      "Epoch 617: 1.1561201349749786 2.077364593638815\n",
      "Epoch 618: 1.1479035736525844 2.008954274865795\n",
      "Epoch 619: 1.144524703842288 1.980690356318616\n",
      "Epoch 620: 1.136992737736608 1.920267931915191\n",
      "Epoch 621: 1.1449055065162845 2.0256836009563326\n",
      "Epoch 622: 1.1369548617973584 1.9139524101299534\n",
      "Epoch 623: 1.1394590802641777 1.9412799156168985\n",
      "Epoch 624: 1.1440634230707285 1.9759853212911\n",
      "Epoch 625: 1.1377341050508412 1.8056331460596537\n",
      "Epoch 626: 1.154621556420888 2.0517240402014973\n",
      "Epoch 627: 1.1437171562739505 1.9261258346472023\n",
      "Epoch 628: 1.150385307559757 2.027778902322227\n",
      "Epoch 629: 1.1530967199002244 2.0626847385825093\n",
      "Epoch 630: 1.1531172636195939 2.065527734414501\n",
      "Epoch 631: 1.1490742275494856 1.9638324313626334\n",
      "Epoch 632: 1.155315310074967 2.102823402912834\n",
      "Epoch 633: 1.1417601690775259 1.9172810647771976\n",
      "Epoch 634: 1.1510159534567739 2.010997298421388\n",
      "Epoch 635: 1.149005741676144 1.9937017899446772\n",
      "Epoch 636: 1.155222921847631 2.01113908577509\n",
      "Epoch 637: 1.150820535642392 1.9526606615164677\n",
      "Epoch 638: 1.1551409338435366 2.000433609996005\n",
      "Epoch 639: 1.1578834103979951 2.0468988071021372\n",
      "Epoch 640: 1.1565305856473265 2.0128183635846515\n",
      "Epoch 641: 1.1591606428192935 2.01835623644454\n",
      "Epoch 642: 1.1569524851263293 2.045253459625615\n",
      "Epoch 643: 1.1514787531975645 1.9551687657003172\n",
      "Epoch 644: 1.163649621435769 2.0838699038768436\n",
      "Epoch 645: 1.1472411948456802 1.9178902458696894\n",
      "Epoch 646: 1.1447468750038605 1.8467790295194404\n",
      "Epoch 647: 1.165686895471076 2.1111351565553234\n",
      "Epoch 648: 1.1576923239875891 1.9899596857567805\n",
      "Epoch 649: 1.161485272855309 2.025151149884347\n",
      "Epoch 650: 1.1597597440083867 1.908835408981916\n",
      "Epoch 651: 1.1650400581070193 2.0420247761455457\n",
      "Epoch 652: 1.1675470577075604 2.062979899616677\n",
      "Epoch 653: 1.159614412837413 1.994903243221295\n",
      "Epoch 654: 1.1582505049578176 1.9941648150426208\n",
      "Epoch 655: 1.1634176578590663 2.057592596278545\n",
      "Epoch 656: 1.1611885297523123 2.007311825230074\n",
      "Epoch 657: 1.154168506409637 1.9096176697993068\n",
      "Epoch 658: 1.1658267998253125 2.0735413463909116\n",
      "Epoch 659: 1.1622960295547577 1.9920497664560168\n",
      "Epoch 660: 1.1605113725197804 1.9769764654165227\n",
      "Epoch 661: 1.1688240775030034 2.0720466891643783\n",
      "Epoch 662: 1.166474635389868 2.043905242100123\n",
      "Epoch 663: 1.1561023168581162 1.891245447819258\n",
      "Epoch 664: 1.170902381424022 2.0453216165097943\n",
      "Epoch 665: 1.1735445476514155 2.0428657951344187\n",
      "Epoch 666: 1.172054013314705 2.0262399185341877\n",
      "Epoch 667: 1.1708034841598174 2.0146171031004765\n",
      "Epoch 668: 1.1671142915453008 1.9348631427426084\n",
      "Epoch 669: 1.1775445636369115 2.0477901564859926\n",
      "Epoch 670: 1.1644419422264496 1.919474895221312\n",
      "Epoch 671: 1.1741559338165275 2.074967660671925\n",
      "Epoch 672: 1.1731059472314012 2.0687478375680235\n",
      "Epoch 673: 1.1625465689120127 1.9400170806618209\n",
      "Epoch 674: 1.1621233434513776 1.9331084122401405\n",
      "Epoch 675: 1.172690681588269 2.078074424231667\n",
      "Epoch 676: 1.1673711696972886 2.0179073495059945\n",
      "Epoch 677: 1.1706102870624606 2.0697200328933634\n",
      "Epoch 678: 1.1703580494763104 1.9874324564763488\n",
      "Epoch 679: 1.1619811693165076 1.8821263438324085\n",
      "Epoch 680: 1.1760639541364406 2.0506844372123743\n",
      "Epoch 681: 1.1768429181245823 2.0749670618770857\n",
      "Epoch 682: 1.1636128240680195 1.8879778704396915\n",
      "Epoch 683: 1.1709499629950786 1.9835964000124204\n",
      "Epoch 684: 1.176966935674532 2.0451598736206993\n",
      "Epoch 685: 1.1748268450232888 2.010486718394343\n",
      "Epoch 686: 1.172105455781458 1.9888604323678445\n",
      "Epoch 687: 1.1773147388707024 2.0552232315545726\n",
      "Epoch 688: 1.174981523032786 2.025749691408062\n",
      "Epoch 689: 1.175802764810576 2.0272108047081865\n",
      "Epoch 690: 1.1741388731261873 2.0438495359399673\n",
      "Epoch 691: 1.1645637415696226 1.914852964763105\n",
      "Epoch 692: 1.1775437201929273 2.0837327430593087\n",
      "Epoch 693: 1.171021545689541 2.0380823234432075\n",
      "Epoch 694: 1.1720299111368093 2.077853785353434\n",
      "Epoch 695: 1.166607405867305 2.0266663943476817\n",
      "Epoch 696: 1.1616206086636185 2.0041200728231723\n",
      "Epoch 697: 1.1650256609464475 2.0303823418621296\n",
      "Epoch 698: 1.164814414701397 1.9838895901176739\n",
      "Epoch 699: 1.1743731417049934 2.121649344069409\n",
      "Epoch 700: 1.1729325183302142 2.090510587257664\n",
      "Epoch 701: 1.167460302805175 1.9853618003596634\n",
      "Epoch 702: 1.171378151734284 2.053763389219159\n",
      "Epoch 703: 1.1702239862054735 2.055377866431336\n",
      "Epoch 704: 1.1655974083142302 1.99677623915332\n",
      "Epoch 705: 1.1637079902216065 1.9796444231160475\n",
      "Epoch 706: 1.1657164709848369 1.9887143991472445\n",
      "Epoch 707: 1.1675292832336142 2.0241719840109442\n",
      "Epoch 708: 1.1698232737370793 2.035350913839469\n",
      "Epoch 709: 1.1734883786457513 2.058245749818177\n",
      "Epoch 710: 1.178176428028091 2.0956221073814563\n",
      "Epoch 711: 1.1707212472550819 2.0256728162977207\n",
      "Epoch 712: 1.1575486624890918 1.8194403182077832\n",
      "Epoch 713: 1.177523685965058 2.0673595592156992\n",
      "Epoch 714: 1.1747159994855776 2.043155876208187\n",
      "Epoch 715: 1.1710098799379844 1.9762692647653097\n",
      "Epoch 716: 1.1640454907921152 1.8903064571396975\n",
      "Epoch 717: 1.1762622377686232 2.0574860911681907\n",
      "Epoch 718: 1.1787825645581391 2.002981652644582\n",
      "Epoch 719: 1.1760841297560203 1.9935265856839561\n",
      "Epoch 720: 1.1809428938255737 2.0650923158709085\n",
      "Epoch 721: 1.1765205975787607 1.9578502910758795\n",
      "Epoch 722: 1.1774642175778434 2.019709090157791\n",
      "Epoch 723: 1.1680699833280164 1.898312445095025\n",
      "Epoch 724: 1.1761840770108003 2.0462077852453637\n",
      "Epoch 725: 1.1718479222055256 2.0058071438445033\n",
      "Epoch 726: 1.1664817877999039 1.9096610999007464\n",
      "Epoch 727: 1.1724817886067012 1.9513913121667488\n",
      "Epoch 728: 1.177878687532817 2.0048043751218154\n",
      "Epoch 729: 1.1805702427200604 2.038253110489647\n",
      "Epoch 730: 1.1731762805250165 1.9773643250294353\n",
      "Epoch 731: 1.1719481861256338 1.9681530248838734\n",
      "Epoch 732: 1.170958734170868 1.931736000361683\n",
      "Epoch 733: 1.1809837229509716 2.076098698308606\n",
      "Epoch 734: 1.1814920816011996 2.1008648023647707\n",
      "Epoch 735: 1.1739589961909729 1.9850473133869209\n",
      "Epoch 736: 1.175209416452425 2.010759548096611\n",
      "Epoch 737: 1.17394430015845 1.980847883428966\n",
      "Epoch 738: 1.1742166094964246 1.9785416003352714\n",
      "Epoch 739: 1.176104874697144 1.9923927012366787\n",
      "Epoch 740: 1.1713424931242489 1.8749667813350874\n",
      "Epoch 741: 1.179252499460744 1.9830882619410506\n",
      "Epoch 742: 1.1719552117063976 1.8948350185906075\n",
      "Epoch 743: 1.1823893626823063 2.0678006992277806\n",
      "Epoch 744: 1.1780966097356345 1.998036497360876\n",
      "Epoch 745: 1.179165780753876 2.004630972509234\n",
      "Epoch 746: 1.1835778630200546 2.0950285707921648\n",
      "Epoch 747: 1.1766786813161292 2.044100782536718\n",
      "Epoch 748: 1.160163503077738 1.9177065378896447\n",
      "Epoch 749: 1.1675883035849126 2.014212936737423\n",
      "Epoch 750: 1.1634300571662926 1.9721611100038532\n",
      "Epoch 751: 1.1715877858447237 2.043941039719496\n",
      "Epoch 752: 1.1584171500053826 1.8746409041865049\n",
      "Epoch 753: 1.1763127461801561 2.0918118518975723\n",
      "Epoch 754: 1.1659716692903694 1.9742224619202564\n",
      "Epoch 755: 1.1646604634177344 1.9361122857977011\n",
      "Epoch 756: 1.160620833248367 1.8662711354233066\n",
      "Epoch 757: 1.1661982405581681 1.968819352593975\n",
      "Epoch 758: 1.1629045745496742 1.9550035485315502\n",
      "Epoch 759: 1.1647548221153303 1.9730341239745295\n",
      "Epoch 760: 1.1663056539956702 1.965890594274193\n",
      "Epoch 761: 1.1779151897816287 2.081608721625768\n",
      "Epoch 762: 1.1693962316786333 1.941286889886656\n",
      "Epoch 763: 1.1805267074161756 2.0743145136767094\n",
      "Epoch 764: 1.1800985705084266 2.0668825274432767\n",
      "Epoch 765: 1.1726861160529571 1.956012847673854\n",
      "Epoch 766: 1.1877647813139627 2.083771880721844\n",
      "Epoch 767: 1.1759959529950046 1.9502345357584432\n",
      "Epoch 768: 1.180106600845589 2.030203465317662\n",
      "Epoch 769: 1.1785547996239625 2.0075038810866097\n",
      "Epoch 770: 1.1721198204166714 1.9509697428767063\n",
      "Epoch 771: 1.1748246698164564 1.9928669913981039\n",
      "Epoch 772: 1.1730788942530228 1.9950697028851319\n",
      "Epoch 773: 1.175246930868462 1.9668602729713875\n",
      "Epoch 774: 1.1734417451383476 1.9514775324234275\n",
      "Epoch 775: 1.1865747041781565 2.098993134805362\n",
      "Epoch 776: 1.188501736263222 2.0506954757968603\n",
      "Epoch 777: 1.182852656139179 1.9646481114324323\n",
      "Epoch 778: 1.185967165253824 2.013231001970683\n",
      "Epoch 779: 1.1844395490522908 2.0910341881077166\n",
      "Epoch 780: 1.1737915821547968 1.9646164138616604\n",
      "Epoch 781: 1.191758869935296 2.18962453706787\n",
      "Epoch 782: 1.1831959356758552 2.1198312684358753\n",
      "Epoch 783: 1.1723311593442 1.9859166627501075\n",
      "Epoch 784: 1.1744433846975588 2.019754884079452\n",
      "Epoch 785: 1.1719566576658207 2.019318801231734\n",
      "Epoch 786: 1.1710418213323723 1.988913309003219\n",
      "Epoch 787: 1.1620316669598556 1.8707335762957873\n",
      "Epoch 788: 1.1703379354775691 1.997393590357857\n",
      "Epoch 789: 1.1719740907252663 2.025390894214763\n",
      "Epoch 790: 1.1751951340536277 2.073895434476763\n",
      "Epoch 791: 1.1649917139926171 1.9780477435131272\n",
      "Epoch 792: 1.1703869198290446 2.025718385462993\n",
      "Epoch 793: 1.1702765273424296 2.0614739778898983\n",
      "Epoch 794: 1.1625825879514962 1.9226274177688791\n",
      "Epoch 795: 1.1655086323726378 1.9208708143722197\n",
      "Epoch 796: 1.1691174033705247 1.9006829597676416\n",
      "Epoch 797: 1.1729928727329737 1.9640889316484784\n",
      "Epoch 798: 1.1778463811902438 1.9921573108254367\n",
      "Epoch 799: 1.1773514612461151 1.9667117251989434\n",
      "Epoch 800: 1.177124613518903 2.0343926448254734\n",
      "Epoch 801: 1.1749925961056273 1.985964658981048\n",
      "Epoch 802: 1.179957250611606 2.047084039569667\n",
      "Epoch 803: 1.178775628421662 2.0210086054432637\n",
      "Epoch 804: 1.175220208981832 2.02350131862926\n",
      "Epoch 805: 1.1673621446963132 1.960305978293802\n",
      "Epoch 806: 1.1791041023504112 2.1120544016014757\n",
      "Epoch 807: 1.178017568433982 2.0771402984542813\n",
      "Epoch 808: 1.165647287412458 1.9602070874952493\n",
      "Epoch 809: 1.1581522666958617 1.9120867281927372\n",
      "Epoch 810: 1.1574695893847933 1.9553021737969036\n",
      "Epoch 811: 1.1532611898613325 1.9119749320993886\n",
      "Epoch 812: 1.1614700136038754 1.98716206728456\n",
      "Epoch 813: 1.1671994116082518 2.0275299821752375\n",
      "Epoch 814: 1.1666677665583243 2.0417365532450047\n",
      "Epoch 815: 1.1697040029415398 2.0734529891667592\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 816: 1.1686166547942265 2.0558238071783412\n",
      "Epoch 817: 1.1601948873090822 2.010241210464497\n",
      "Epoch 818: 1.150618385029723 1.9258408855428137\n",
      "Epoch 819: 1.1636429174708893 2.0233682903580497\n",
      "Epoch 820: 1.1622884142555794 2.0014636002126447\n",
      "Epoch 821: 1.1555748556733847 1.906875567540085\n",
      "Epoch 822: 1.1664335200140223 2.0571278055304023\n",
      "Epoch 823: 1.1583224588173926 1.9869757506956676\n",
      "Epoch 824: 1.1702089173813757 2.1543993462777062\n",
      "Epoch 825: 1.1616518864819756 2.0773275184740485\n",
      "Epoch 826: 1.1551967684697095 2.0063270888888423\n",
      "Epoch 827: 1.152287628406903 1.9779246087420448\n",
      "Epoch 828: 1.157850677359235 2.057287178586857\n",
      "Epoch 829: 1.1556715647828006 2.0239885485233216\n",
      "Epoch 830: 1.1565890096397193 2.0503171254560946\n",
      "Epoch 831: 1.1595389084512515 1.98795193268577\n",
      "Epoch 832: 1.1589800571798163 2.011033267696134\n",
      "Epoch 833: 1.1556691202822504 1.9693737537898452\n",
      "Epoch 834: 1.157665567164785 2.0208205094235487\n",
      "Epoch 835: 1.1595353260212122 2.063046521990212\n",
      "Epoch 836: 1.1557062861486722 1.9994191106493338\n",
      "Epoch 837: 1.1565809689923208 1.983921711149201\n",
      "Epoch 838: 1.1623608098363054 2.0646902809727394\n",
      "Epoch 839: 1.1581531743891427 1.987295732082542\n",
      "Epoch 840: 1.1590402311588939 1.9689879988643828\n",
      "Epoch 841: 1.1646506455863843 2.0480950713908213\n",
      "Epoch 842: 1.162215740353453 1.9799758074439537\n",
      "Epoch 843: 1.1571640072959855 1.921655076066351\n",
      "Epoch 844: 1.1580560500600952 1.9277735300278482\n",
      "Epoch 845: 1.1649186113385384 1.9996409990007527\n",
      "Epoch 846: 1.1646426722474732 2.003677002909093\n",
      "Epoch 847: 1.163233802067977 1.9591389964503692\n",
      "Epoch 848: 1.1707356847623558 2.063112166983802\n",
      "Epoch 849: 1.1649031859386627 1.9707133701977613\n",
      "Epoch 850: 1.1669768604197768 2.003781069027834\n",
      "Epoch 851: 1.1666918279123084 1.9670802100651719\n",
      "Epoch 852: 1.177111771987998 2.082291134554474\n",
      "Epoch 853: 1.1714490445121664 1.9742670079602334\n",
      "Epoch 854: 1.1690847946719853 1.9717368670033464\n",
      "Epoch 855: 1.1704196198091974 2.001142604751988\n",
      "Epoch 856: 1.1643601918228275 1.9346720864323468\n",
      "Epoch 857: 1.169987727850821 1.9859981783905607\n",
      "Epoch 858: 1.1679002555552322 1.9558523607403306\n",
      "Epoch 859: 1.1700790689829543 1.9659696296969515\n",
      "Epoch 860: 1.1702633562751767 2.035529723782505\n",
      "Epoch 861: 1.1668381815422824 1.9730851200413737\n",
      "Epoch 862: 1.1710780627357755 2.0021502399044393\n",
      "Epoch 863: 1.1778865245313674 2.0584634789808436\n",
      "Epoch 864: 1.1706303717546014 1.9793857058274713\n",
      "Epoch 865: 1.1720638650705177 2.002020048112528\n",
      "Epoch 866: 1.177027344548512 2.094128393383114\n",
      "Epoch 867: 1.1590711620519718 1.930179604717959\n",
      "Epoch 868: 1.1595880558962133 1.9178769953065604\n",
      "Epoch 869: 1.163754641289815 1.9709385703174107\n",
      "Epoch 870: 1.1626672733179608 1.9655228206886322\n",
      "Epoch 871: 1.1630797077085382 2.0002004967166447\n",
      "Epoch 872: 1.1660986739522754 2.071909935186307\n",
      "Epoch 873: 1.160184645422692 1.9995763508452455\n",
      "Epoch 874: 1.160804198929034 2.0300445505981894\n",
      "Epoch 875: 1.1585661002935455 2.0156828588956004\n",
      "Epoch 876: 1.1608319160411553 2.0008201812945963\n",
      "Epoch 877: 1.1675099494614467 2.11487769035713\n",
      "Epoch 878: 1.1590479574905752 2.005462536710277\n",
      "Epoch 879: 1.1599498451300045 2.0019678492327615\n",
      "Epoch 880: 1.1551786199453369 2.0250694826853666\n",
      "Epoch 881: 1.1446999750255613 1.9146614964687192\n",
      "Epoch 882: 1.1607304073807792 2.100712842396314\n",
      "Epoch 883: 1.1587528035888388 2.075430378710201\n",
      "Epoch 884: 1.1450873628445297 1.9458641513033033\n",
      "Epoch 885: 1.1516276042155942 1.9869954395285911\n",
      "Epoch 886: 1.1558943781481392 1.9879195865884405\n",
      "Epoch 887: 1.1588996968992071 2.0238636993884156\n",
      "Epoch 888: 1.1589243138387106 1.9685674227858476\n",
      "Epoch 889: 1.1574874914236424 1.9914169005594886\n",
      "Epoch 890: 1.1458185129034988 1.8453262503642136\n",
      "Epoch 891: 1.149657736705517 1.9360649678219166\n",
      "Epoch 892: 1.1525349621243426 2.0003436914457113\n",
      "Epoch 893: 1.1566643190248354 2.0899291699728164\n",
      "Epoch 894: 1.1538013519485895 2.0285061124476362\n",
      "Epoch 895: 1.1494894712829409 1.9841138725718106\n",
      "Epoch 896: 1.1467280116705567 1.9875843391309578\n",
      "Epoch 897: 1.1457867423798365 1.965955036334711\n",
      "Epoch 898: 1.1500922673760217 2.0171676734984088\n",
      "Epoch 899: 1.1461515369907496 1.9924543962760368\n",
      "Epoch 900: 1.1561367888381597 2.124404315519627\n",
      "Epoch 901: 1.143623481118157 1.9879421847563254\n",
      "Epoch 902: 1.1371309103687557 1.9300487974583553\n",
      "Epoch 903: 1.1392961111037059 1.923134522407426\n",
      "Epoch 904: 1.1407954023431104 1.9720533933578297\n",
      "Epoch 905: 1.1447308147935196 1.9998299887445616\n",
      "Epoch 906: 1.1464585126021198 2.0148125135693653\n",
      "Epoch 907: 1.1404403740371 1.9297813616045891\n",
      "Epoch 908: 1.1463488767108843 2.0096770631354377\n",
      "Epoch 909: 1.1546029622929534 2.0832874132502397\n",
      "Epoch 910: 1.150169572833268 2.03405456699914\n",
      "Epoch 911: 1.1461421224307 1.940212083531217\n",
      "Epoch 912: 1.1461235515941184 1.9879469239195826\n",
      "Epoch 913: 1.1437277165448931 1.9458425306722948\n",
      "Epoch 914: 1.1465826613349566 1.963993027082908\n",
      "Epoch 915: 1.153700095135566 2.061082796293308\n",
      "Epoch 916: 1.145535162769503 1.9399497199454492\n",
      "Epoch 917: 1.1514181013196756 2.00126064285679\n",
      "Epoch 918: 1.145316405324972 1.8983230429796274\n",
      "Epoch 919: 1.14742180546 1.9248133926280373\n",
      "Epoch 920: 1.1453945133821228 1.9258806454553379\n",
      "Epoch 921: 1.1564822331850053 2.0417494917942447\n",
      "Epoch 922: 1.16532666842355 2.111873507098976\n",
      "Epoch 923: 1.1540591309061792 1.9727663628126484\n",
      "Epoch 924: 1.1511551928026063 1.9398995635622858\n",
      "Epoch 925: 1.153896463580002 1.958796187945312\n",
      "Epoch 926: 1.1585792016906362 1.9770359681201741\n",
      "Epoch 927: 1.1638372257027831 1.987788305134844\n",
      "Epoch 928: 1.1693475765031072 2.0466449244121923\n",
      "Epoch 929: 1.1613297289146602 1.9509056905792623\n",
      "Epoch 930: 1.1602256163595408 1.907791824877854\n",
      "Epoch 931: 1.1762509516408628 2.111297411782571\n",
      "Epoch 932: 1.1699169121556179 2.0205901983640597\n",
      "Epoch 933: 1.1684207492079968 1.9272494985274782\n",
      "Epoch 934: 1.1775275555358624 2.0234431331006006\n",
      "Epoch 935: 1.1835057205367496 2.096536323344183\n",
      "Epoch 936: 1.1808714871626944 2.06793651984597\n",
      "Epoch 937: 1.1711764282577117 1.944775568850033\n",
      "Epoch 938: 1.1765847908624938 2.059669845699896\n",
      "Epoch 939: 1.1689618571823044 1.99094115269792\n",
      "Epoch 940: 1.1702445180640104 2.0274149394537373\n",
      "Epoch 941: 1.1673014771928076 2.006856804873804\n",
      "Epoch 942: 1.1680741730951625 2.0138214259752067\n",
      "Epoch 943: 1.1632470351155013 1.9746100270936742\n",
      "Epoch 944: 1.164775740105387 1.9920699533172106\n",
      "Epoch 945: 1.1688593880528286 2.0088814901064453\n",
      "Epoch 946: 1.1747648520414604 2.064169474455046\n",
      "Epoch 947: 1.166189301475967 1.9512623293150524\n",
      "Epoch 948: 1.1723293093984961 1.9991619097278583\n",
      "Epoch 949: 1.1724157247810891 2.0119349303671377\n",
      "Epoch 950: 1.1749165447698378 2.056822162737071\n",
      "Epoch 951: 1.1736709326235968 2.0236610959914625\n",
      "Epoch 952: 1.1750017021555834 2.0385020828048606\n",
      "Epoch 953: 1.1767926748338744 2.078712474345494\n",
      "Epoch 954: 1.1712354911892442 1.988543257494375\n",
      "Epoch 955: 1.1740383287480203 2.053345404770091\n",
      "Epoch 956: 1.1751871446465831 2.0933373481889297\n",
      "Epoch 957: 1.1641647182469066 1.9155137587626623\n",
      "Epoch 958: 1.1691004333247237 1.9414331144327648\n",
      "Epoch 959: 1.1772536725425993 2.067633262103231\n",
      "Epoch 960: 1.1779986870370482 2.086082028772622\n",
      "Epoch 961: 1.1738897563079478 2.047915322270108\n",
      "Epoch 962: 1.1619161143374852 1.8923934588954494\n",
      "Epoch 963: 1.1697457741957606 1.9844759110197736\n",
      "Epoch 964: 1.1697351096822857 1.9741557080092809\n",
      "Epoch 965: 1.1674910142105261 1.970702981080298\n",
      "Epoch 966: 1.1708913097774833 2.035510742775975\n",
      "Epoch 967: 1.163560206727453 1.959034424532293\n",
      "Epoch 968: 1.1645150122282628 2.0045453242661675\n",
      "Epoch 969: 1.1679107423551707 2.063905013621921\n",
      "Epoch 970: 1.1677195713875725 2.0433237073388213\n",
      "Epoch 971: 1.17002548380566 2.087239453154267\n",
      "Epoch 972: 1.1669677656803277 2.00776004658578\n",
      "Epoch 973: 1.1693951222961703 2.0054310660716648\n",
      "Epoch 974: 1.163591461957682 1.9957651646191144\n",
      "Epoch 975: 1.1631281436776817 1.996814233291856\n",
      "Epoch 976: 1.1657782005156223 2.0337180441688014\n",
      "Epoch 977: 1.1613277185536075 1.9761369889578149\n",
      "Epoch 978: 1.1715340660463078 2.0889472256757897\n",
      "Epoch 979: 1.1643274955849614 2.0325757476315434\n",
      "Epoch 980: 1.169053706321655 2.1097168551532635\n",
      "Epoch 981: 1.162086973500796 1.9555235042178425\n",
      "Epoch 982: 1.1659416789106982 2.0106789994605276\n",
      "Epoch 983: 1.169102413154393 2.0281705562256747\n",
      "Epoch 984: 1.1554108232659615 1.9228184709910991\n",
      "Epoch 985: 1.1588930741118912 1.921251824994387\n",
      "Epoch 986: 1.1550587111137056 1.8840277756684038\n",
      "Epoch 987: 1.1638976152823144 2.025125792607056\n",
      "Epoch 988: 1.1671422605177448 2.097307271135845\n",
      "Epoch 989: 1.1556821951123852 1.9115550838186852\n",
      "Epoch 990: 1.160829285965992 1.940623122785074\n",
      "Epoch 991: 1.165446317556802 2.000186893753661\n",
      "Epoch 992: 1.1718972107106718 2.076132788705562\n",
      "Epoch 993: 1.163709861587923 1.9829789758500083\n",
      "Epoch 994: 1.1624906108985786 2.000615035032854\n",
      "Epoch 995: 1.1613760832770657 1.9794077021416954\n",
      "Epoch 996: 1.1678270786062839 2.0712825199502247\n",
      "Epoch 997: 1.1632112053389412 2.0243115044550732\n",
      "Epoch 998: 1.1598619426016632 1.9867915874642743\n",
      "Epoch 999: 1.1667041528951323 2.064176860885111\n"
     ]
    }
   ],
   "source": [
    "epochs = 1000\n",
    "mini_batch_size = 10\n",
    "eta = 0.01/mini_batch_size\n",
    "\n",
    "a = 3.\n",
    "b = 3.\n",
    "def update_mini_batch(mini_batch, eta):\n",
    "    global a, b\n",
    "    a0 = a\n",
    "    b0 = b\n",
    "    for x, y, in mini_batch:\n",
    "        e = eta*(a0+b0*x-y)\n",
    "        a -= e\n",
    "        b -= x*e\n",
    "    \n",
    "training_data = list(zip(x_data,y_data))\n",
    "for j in range(epochs):\n",
    "    np.random.shuffle(training_data)\n",
    "    mini_batches = [training_data[k:k+mini_batch_size]\n",
    "                    for k in range(0, len(training_data), mini_batch_size)]\n",
    "    for mini_batch in mini_batches:\n",
    "        update_mini_batch(mini_batch, eta)\n",
    "    print (\"Epoch {0}: {1} {2}\".format(j,a,b))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Challenge 14.2\n",
    "\n",
    "Use SGD to train the single neuron in the previous notebook using a linearly separable set of 100 points, divided by the line $-\\frac{5}{2}x+\\frac{3}{2}y+3=0$ "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "### We provide a set of randomly generated training points \n",
    "num_points = 100\n",
    "w1 = -2.5\n",
    "w2 = 1.5\n",
    "w0 = 3.\n",
    "np.random.seed(637163) # we make sure we always generate the same sequence\n",
    "x_data = np.random.rand(num_points)*10.\n",
    "y_data = np.random.rand(num_points)*10.\n",
    "z_data = np.zeros(num_points)\n",
    "for i in range(len(z_data)):\n",
    "    if (y_data[i] > (-w0-w1*x_data[i])/w2):\n",
    "        z_data[i] = 1.\n",
    "\n",
    "pyplot.scatter(x_data,y_data,c=z_data,marker='o',linewidth=1.5,edgecolors='black')\n",
    "pyplot.plot(x_data,(-w1*x_data-w0)/w2)\n",
    "pyplot.gray()\n",
    "pyplot.xlim(0,10)\n",
    "pyplot.ylim(0,10);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You will need the following auxiliary functions:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "def sigmoid(z):\n",
    "    \"\"\"The sigmoid function.\"\"\"\n",
    "    return 1.0/(1.0+np.exp(-z))\n",
    "\n",
    "def sigmoid_prime(z):\n",
    "    \"\"\"Derivative of the sigmoid function.\"\"\"\n",
    "    return sigmoid(z)*(1-sigmoid(z))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## A simple network to classify handwritten digits\n",
    "Most of this section has been taken from M. Nielsen's free on-line book: \"Neural Networks and Deep Learning\" http://neuralnetworksanddeeplearning.com/\n",
    "\n",
    "In this section we discuss a neural network which can solve the more interesting and difficult problem, namely, recognizing individual handwritten digits.\n",
    "\n",
    "The input layer of the network contains neurons encoding the values of the input pixels. Our training data for the network will consist of many 28 by 28\n",
    " pixel images of scanned handwritten digits, and so the input layer contains 784=28×28\n",
    " neurons. The input pixels are greyscale, with a value of 0.0\n",
    " representing white, a value of 1.0\n",
    "representing black, and in between values representing gradually darkening shades of grey.\n",
    "\n",
    "The second layer of the network is a hidden layer. We denote the number of neurons in this hidden layer by $n$\n",
    ", and we'll experiment with different values for $n$\n",
    ". The example shown illustrates a small hidden layer, containing just $n=15$\n",
    " neurons.\n",
    " \n",
    " The output layer of the network contains 10 neurons. If the first neuron fires, i.e., has an output $\\sim 1$\n",
    ", then that will indicate that the network thinks the digit is a 0\n",
    ". If the second neuron fires then that will indicate that the network thinks the digit is a 1\n",
    ". And so on. A little more precisely, we number the output neurons from 0\n",
    " through 9\n",
    ", and figure out which neuron has the highest activation value. If that neuron is, say, neuron number 6\n",
    ", then our network will guess that the input digit was a 6\n",
    ". And so on for the other output neurons.\n",
    "\n",
    "<img src=\"figures/nnetwork.png\" style=\"width: 500px;\"/>\n",
    "#### Network to identify single digits. The output layer has 10 neurons, one for each digit.\n",
    "\n",
    "\n",
    "\n",
    "The first thing we'll need is a data set to learn from - a so-called training data set. We'll use the MNIST data set, which contains tens of thousands of scanned images of handwritten digits, together with their correct classifications. MNIST's name comes from the fact that it is a modified subset of two data sets collected by NIST, the United States' National Institute of Standards and Technology. Here's a few images from MNIST:\n",
    "\n",
    "<img src=\"figures/digits_separate.png\" style=\"width: 250px;\"/>\n",
    "\n",
    "\n",
    "\n",
    "The MNIST data comes in two parts. The first part contains 60,000 images to be used as training data. These images are scanned handwriting samples from 250 people, half of whom were US Census Bureau employees, and half of whom were high school students. The images are greyscale and 28 by 28 pixels in size. The second part of the MNIST data set is 10,000 images to be used as test data. Again, these are 28 by 28 greyscale images. We'll use the test data to evaluate how well our neural network has learned to recognize digits. To make this a good test of performance, the test data was taken from a different set of 250 people than the original training data (albeit still a group split between Census Bureau employees and high school students). This helps give us confidence that our system can recognize digits from people whose writing it didn't see during training.\n",
    "\n",
    "In practice, we are going to split the data a little differently. We'll leave the test images as is, but split the 60,000-image MNIST training set into two parts: a set of 50,000 images, which we'll use to train our neural network, and a separate 10,000 image validation set.\n",
    "\n",
    "We'll use the notation $x$\n",
    " to denote a training input. It'll be convenient to regard each training input $x$\n",
    " as a 28×28=784-dimensional vector. Each entry in the vector represents the grey value for a single pixel in the image. We'll denote the corresponding desired output by y=y(x)\n",
    ", where y\n",
    " is a 10\n",
    "-dimensional vector. For example, if a particular training image, $x$\n",
    ", depicts a 6\n",
    ", then $y(x)=(0,0,0,0,0,0,1,0,0,0)^T$\n",
    " is the desired output from the network. Note that T\n",
    " here is the transpose operation, turning a row vector into an ordinary (column) vector.\n",
    " "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "mnist_loader\n",
    "~~~~~~~~~~~~\n",
    "\n",
    "A library to load the MNIST image data.  For details of the data\n",
    "structures that are returned, see the doc strings for ``load_data``\n",
    "and ``load_data_wrapper``.  In practice, ``load_data_wrapper`` is the\n",
    "function usually called by our neural network code.\n",
    "\"\"\"\n",
    "\n",
    "#### Libraries\n",
    "# Standard library\n",
    "import pickle\n",
    "import gzip\n",
    "\n",
    "# Third-party libraries\n",
    "import numpy as np\n",
    "\n",
    "def load_data():\n",
    "    \"\"\"Return the MNIST data as a tuple containing the training data,\n",
    "    the validation data, and the test data.\n",
    "\n",
    "    The ``training_data`` is returned as a tuple with two entries.\n",
    "    The first entry contains the actual training images.  This is a\n",
    "    numpy ndarray with 50,000 entries.  Each entry is, in turn, a\n",
    "    numpy ndarray with 784 values, representing the 28 * 28 = 784\n",
    "    pixels in a single MNIST image.\n",
    "\n",
    "    The second entry in the ``training_data`` tuple is a numpy ndarray\n",
    "    containing 50,000 entries.  Those entries are just the digit\n",
    "    values (0...9) for the corresponding images contained in the first\n",
    "    entry of the tuple.\n",
    "\n",
    "    The ``validation_data`` and ``test_data`` are similar, except\n",
    "    each contains only 10,000 images.\n",
    "\n",
    "    This is a nice data format, but for use in neural networks it's\n",
    "    helpful to modify the format of the ``training_data`` a little.\n",
    "    That's done in the wrapper function ``load_data_wrapper()``, see\n",
    "    below.\n",
    "    \"\"\"\n",
    "    f = gzip.open('data/mnist.pkl.gz', 'rb')\n",
    "    training_data, validation_data, test_data = pickle.load(f, encoding='latin1')\n",
    "    f.close()\n",
    "    return (training_data, validation_data, test_data)\n",
    "\n",
    "def load_data_wrapper():\n",
    "    \"\"\"Return a tuple containing ``(training_data, validation_data,\n",
    "    test_data)``. Based on ``load_data``, but the format is more\n",
    "    convenient for use in our implementation of neural networks.\n",
    "\n",
    "    In particular, ``training_data`` is a list containing 50,000\n",
    "    2-tuples ``(x, y)``.  ``x`` is a 784-dimensional numpy.ndarray\n",
    "    containing the input image.  ``y`` is a 10-dimensional\n",
    "    numpy.ndarray representing the unit vector corresponding to the\n",
    "    correct digit for ``x``.\n",
    "\n",
    "    ``validation_data`` and ``test_data`` are lists containing 10,000\n",
    "    2-tuples ``(x, y)``.  In each case, ``x`` is a 784-dimensional\n",
    "    numpy.ndarry containing the input image, and ``y`` is the\n",
    "    corresponding classification, i.e., the digit values (integers)\n",
    "    corresponding to ``x``.\n",
    "\n",
    "    Obviously, this means we're using slightly different formats for\n",
    "    the training data and the validation / test data.  These formats\n",
    "    turn out to be the most convenient for use in our neural network\n",
    "    code.\"\"\"\n",
    "    tr_d, va_d, te_d = load_data()\n",
    "    training_inputs = [np.reshape(x, (784, 1)) for x in tr_d[0]]\n",
    "    training_results = [vectorized_result(y) for y in tr_d[1]]\n",
    "    training_data = list(zip(training_inputs, training_results))\n",
    "    validation_inputs = [np.reshape(x, (784, 1)) for x in va_d[0]]\n",
    "    validation_data = list(zip(validation_inputs, va_d[1]))\n",
    "    test_inputs = [np.reshape(x, (784, 1)) for x in te_d[0]]\n",
    "    test_data = list(zip(test_inputs, te_d[1]))\n",
    "    return (training_data, validation_data, test_data)\n",
    "\n",
    "def vectorized_result(j):\n",
    "    \"\"\"Return a 10-dimensional unit vector with a 1.0 in the jth\n",
    "    position and zeroes elsewhere.  This is used to convert a digit\n",
    "    (0...9) into a corresponding desired output from the neural\n",
    "    network.\"\"\"\n",
    "    e = np.zeros((10, 1))\n",
    "    e[j] = 1.0\n",
    "    return e"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note also that the biases and weights are stored as lists of Numpy matrices. So, for example `net.weights[1]` is a Numpy matrix storing the weights connecting the second and third layers of neurons. (It's not the first and second layers, since Python's list indexing starts at 0.) Since `net.weights[1]` is rather verbose, let's just denote that matrix $w$\n",
    ". It's a matrix such that $w_{jk}$\n",
    " is the weight for the connection between the $k^{th}$\n",
    " neuron in the second layer, and the $j^{th}$\n",
    " neuron in the third layer. This ordering of the $j$\n",
    " and $k$\n",
    " indices may seem strange. The big advantage of using this ordering is that it means that the vector of activations of the third layer of neurons is:\n",
    "$$a'=\\mathrm {sigmoid}(wa+b)$$\n",
    "\n",
    "There's quite a bit going on in this equation, so let's unpack it piece by piece. $a$\n",
    " is the vector of activations of the second layer of neurons. To obtain $a'$\n",
    " we multiply $a$\n",
    " by the weight matrix $w$\n",
    ", and add the vector $b$\n",
    " of biases. We then apply the function `sigmoid`\n",
    "elementwise to every entry in the vector $wa+b$.\n",
    "\n",
    "Of course, the main thing we want our Network objects to do is to learn. To that end we'll give them an SGD method which implements stochastic gradient descent. \n",
    "\n",
    "<!--\n",
    "The training_data is a list of tuples `(x, y)` representing the training inputs and corresponding desired outputs. The variables `epochs` and `mini_batch_size` are what you'd expect - the number of epochs to train for, and the size of the mini-batches to use when sampling. `eta` is the learning rate, $\\eta$. If the optional argument `test_data` is supplied, then the program will evaluate the network after each epoch of training, and print out partial progress. This is useful for tracking progress, but slows things down substantially.\n",
    "\n",
    "The code works as follows. In each epoch, it starts by randomly shuffling the training data, and then partitions it into mini-batches of the appropriate size. This is an easy way of sampling randomly from the training data. Then for each `mini_batch` we apply a single step of gradient descent. This is done by the code `self.update_mini_batch(mini_batch, eta)`, which updates the network weights and biases according to a single iteration of gradient descent, using just the training data in `mini_batch`.\n",
    "-->\n",
    "\n",
    "Most of the work is done by the line\n",
    "```\n",
    "            delta_nabla_b, delta_nabla_w = self.backprop(x, y)\n",
    "```\n",
    "\n",
    "This invokes something called the *backpropagation* algorithm, which is a fast way of computing the gradient of the cost function. So `update_mini_batch` works simply by computing these gradients for every training example in the `mini_batch`, and then updating `self.weights` and `self.biases` appropriately.\n",
    "\n",
    "The activation $a_{lj}$\n",
    " of the $j^{th}$\n",
    " neuron in the $l^{th}$\n",
    "layer is related to the activations in the $(l-1)^{th}$\n",
    " layer by the equation\n",
    "$$a^l_j=\\mathrm{sigmoid}(\\sum_k w_{jk}^l a^{l-1}_k+b^l_j)$$\n",
    "where the sum is over all neurons $k$\n",
    " in the $(l−1)^{th}$\n",
    " layer. To rewrite this expression in a matrix form we define a weight matrix $w^l$\n",
    " for each layer, $l$\n",
    ". The entries of the weight matrix $w^l$\n",
    " are just the weights connecting to the $l^{th}$\n",
    " layer of neurons, that is, the entry in the $j^{th}$\n",
    " row and $k^{th}$\n",
    " column is $w^l_{jk}$. Similarly, for each layer $l$\n",
    " we define a bias vector, $b^l$. You can probably guess how this works - the components of the bias vector are just the values $b^l_j$\n",
    ", one component for each neuron in the $l^{th}$\n",
    " layer. And finally, we define an activation vector $a^l$\n",
    "whose components are the activations $a^l_j$.\n",
    "\n",
    "With these notations in mind, these equations can be rewritten in the beautiful and compact vectorized form\n",
    "$$a^l=\\mathrm{sigmoid}(w^la^{l-1}+b^l).$$\n",
    "This expression gives us a much more global way of thinking about how the activations in one layer relate to activations in the previous layer: we just apply the weight matrix to the activations, then add the bias vector, and finally apply the `sigmoid` function.\n",
    "\n",
    "Apart from `self.backprop` the program is self-explanatory - all the heavy lifting is done in `self.SGD` and `self.update_mini_batch`, which we've already discussed. The `self.backprop` method makes use of a few extra functions to help in computing the gradient, namely sigmoid_prime, which computes the derivative of the sigmoid\n",
    " function, and `self.cost_derivative`. You can get the gist of these (and perhaps the details) just by looking at the code and documentation strings. Note that while the program appears lengthy, much of the code is documentation strings intended to make the code easy to understand. In fact, the program contains just 74 lines of non-whitespace, non-comment code. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "network.py\n",
    "~~~~~~~~~~\n",
    "\n",
    "A module to implement the stochastic gradient descent learning\n",
    "algorithm for a feedforward neural network.  Gradients are calculated\n",
    "using backpropagation.  Note that I have focused on making the code\n",
    "simple, easily readable, and easily modifiable.  It is not optimized,\n",
    "and omits many desirable features.\n",
    "\"\"\"\n",
    "\n",
    "#### Libraries\n",
    "# Standard library\n",
    "import random\n",
    "\n",
    "# Third-party libraries\n",
    "import numpy as np\n",
    "\n",
    "class Network(object):\n",
    "\n",
    "    def __init__(self, sizes):\n",
    "        \"\"\"The list ``sizes`` contains the number of neurons in the\n",
    "        respective layers of the network.  For example, if the list\n",
    "        was [2, 3, 1] then it would be a three-layer network, with the\n",
    "        first layer containing 2 neurons, the second layer 3 neurons,\n",
    "        and the third layer 1 neuron.  The biases and weights for the\n",
    "        network are initialized randomly, using a Gaussian\n",
    "        distribution with mean 0, and variance 1.  Note that the first\n",
    "        layer is assumed to be an input layer, and by convention we\n",
    "        won't set any biases for those neurons, since biases are only\n",
    "        ever used in computing the outputs from later layers.\"\"\"\n",
    "        self.num_layers = len(sizes)\n",
    "        self.sizes = sizes\n",
    "        self.biases = [np.random.randn(y, 1) for y in sizes[1:]]\n",
    "        self.weights = [np.random.randn(y, x)\n",
    "                        for x, y in zip(sizes[:-1], sizes[1:])]\n",
    "\n",
    "    def feedforward(self, a):\n",
    "        \"\"\"Return the output of the network if ``a`` is input.\"\"\"\n",
    "        for b, w in zip(self.biases, self.weights):\n",
    "            a = sigmoid(np.dot(w, a)+b)\n",
    "        return a\n",
    "\n",
    "    def SGD(self, training_data, epochs, mini_batch_size, eta,\n",
    "            test_data=None):\n",
    "        \"\"\"Train the neural network using mini-batch stochastic\n",
    "        gradient descent.  The ``training_data`` is a list of tuples\n",
    "        ``(x, y)`` representing the training inputs and the desired\n",
    "        outputs.  The other non-optional parameters are\n",
    "        self-explanatory.  If ``test_data`` is provided then the\n",
    "        network will be evaluated against the test data after each\n",
    "        epoch, and partial progress printed out.  This is useful for\n",
    "        tracking progress, but slows things down substantially.\"\"\"\n",
    "        if test_data: n_test = len(test_data)\n",
    "        n = len(training_data)\n",
    "        for j in range(epochs):\n",
    "            random.shuffle(training_data)\n",
    "            mini_batches = [\n",
    "                training_data[k:k+mini_batch_size]\n",
    "                for k in range(0, n, mini_batch_size)]\n",
    "            for mini_batch in mini_batches:\n",
    "                self.update_mini_batch(mini_batch, eta)\n",
    "            if test_data:\n",
    "                print (\"Epoch {0}: {1} / {2}\".format(\n",
    "                    j, self.evaluate(test_data), n_test))\n",
    "            else:\n",
    "                print (\"Epoch {0} complete\".format(j))\n",
    "\n",
    "    def update_mini_batch(self, mini_batch, eta):\n",
    "        \"\"\"Update the network's weights and biases by applying\n",
    "        gradient descent using backpropagation to a single mini batch.\n",
    "        The ``mini_batch`` is a list of tuples ``(x, y)``, and ``eta``\n",
    "        is the learning rate.\"\"\"\n",
    "        nabla_b = [np.zeros(b.shape) for b in self.biases]\n",
    "        nabla_w = [np.zeros(w.shape) for w in self.weights]\n",
    "        for x, y in mini_batch:\n",
    "            delta_nabla_b, delta_nabla_w = self.backprop(x, y)\n",
    "            nabla_b = [nb+dnb for nb, dnb in zip(nabla_b, delta_nabla_b)]\n",
    "            nabla_w = [nw+dnw for nw, dnw in zip(nabla_w, delta_nabla_w)]\n",
    "        self.weights = [w-(eta/len(mini_batch))*nw\n",
    "                        for w, nw in zip(self.weights, nabla_w)]\n",
    "        self.biases = [b-(eta/len(mini_batch))*nb\n",
    "                       for b, nb in zip(self.biases, nabla_b)]\n",
    "\n",
    "    def backprop(self, x, y):\n",
    "        \"\"\"Return a tuple ``(nabla_b, nabla_w)`` representing the\n",
    "        gradient for the cost function C_x.  ``nabla_b`` and\n",
    "        ``nabla_w`` are layer-by-layer lists of numpy arrays, similar\n",
    "        to ``self.biases`` and ``self.weights``.\"\"\"\n",
    "        nabla_b = [np.zeros(b.shape) for b in self.biases]\n",
    "        nabla_w = [np.zeros(w.shape) for w in self.weights]\n",
    "        # feedforward\n",
    "        activation = x\n",
    "        activations = [x] # list to store all the activations, layer by layer\n",
    "        zs = [] # list to store all the z vectors, layer by layer\n",
    "        for b, w in zip(self.biases, self.weights):\n",
    "            z = np.dot(w, activation)+b\n",
    "            zs.append(z)\n",
    "            activation = sigmoid(z)\n",
    "            activations.append(activation)\n",
    "        # backward pass\n",
    "        delta = self.cost_derivative(activations[-1], y) * \\\n",
    "            sigmoid_prime(zs[-1])\n",
    "        nabla_b[-1] = delta\n",
    "        nabla_w[-1] = np.dot(delta, activations[-2].transpose())\n",
    "        # Note that the variable l in the loop below is used a little\n",
    "        # differently to the notation in Chapter 2 of the book.  Here,\n",
    "        # l = 1 means the last layer of neurons, l = 2 is the\n",
    "        # second-last layer, and so on.  It's a renumbering of the\n",
    "        # scheme in the book, used here to take advantage of the fact\n",
    "        # that Python can use negative indices in lists.\n",
    "        for l in range(2, self.num_layers):\n",
    "            z = zs[-l]\n",
    "            sp = sigmoid_prime(z)\n",
    "            delta = np.dot(self.weights[-l+1].transpose(), delta) * sp\n",
    "            nabla_b[-l] = delta\n",
    "            nabla_w[-l] = np.dot(delta, activations[-l-1].transpose())\n",
    "        return (nabla_b, nabla_w)\n",
    "\n",
    "    def evaluate(self, test_data):\n",
    "        \"\"\"Return the number of test inputs for which the neural\n",
    "        network outputs the correct result. Note that the neural\n",
    "        network's output is assumed to be the index of whichever\n",
    "        neuron in the final layer has the highest activation.\"\"\"\n",
    "        test_results = [(np.argmax(self.feedforward(x)), y)\n",
    "                        for (x, y) in test_data]\n",
    "        return sum(int(x == y) for (x, y) in test_results)\n",
    "\n",
    "    def cost_derivative(self, output_activations, y):\n",
    "        \"\"\"Return the vector of partial derivatives \\partial C_x /\n",
    "        \\partial a for the output activations.\"\"\"\n",
    "        return (output_activations-y)\n",
    "\n",
    "#### Miscellaneous functions\n",
    "def sigmoid(z):\n",
    "    \"\"\"The sigmoid function.\"\"\"\n",
    "    return 1.0/(1.0+np.exp(-z))\n",
    "\n",
    "def sigmoid_prime(z):\n",
    "    \"\"\"Derivative of the sigmoid function.\"\"\"\n",
    "    return sigmoid(z)*(1-sigmoid(z))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We first load the MNIST data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "training_data, validation_data, test_data = load_data_wrapper()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "After loading the MNIST data, we'll set up a Network with 30 hidden neurons. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "net = Network([784, 30, 10])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, we'll use stochastic gradient descent to learn from the MNIST training_data over 30 epochs, with a mini-batch size of 10, and a learning rate of $\\eta$=3.0:\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0: 9125 / 10000\n",
      "Epoch 1: 9201 / 10000\n",
      "Epoch 2: 9285 / 10000\n",
      "Epoch 3: 9317 / 10000\n",
      "Epoch 4: 9299 / 10000\n",
      "Epoch 5: 9388 / 10000\n",
      "Epoch 6: 9394 / 10000\n",
      "Epoch 7: 9397 / 10000\n",
      "Epoch 8: 9425 / 10000\n",
      "Epoch 9: 9395 / 10000\n",
      "Epoch 10: 9408 / 10000\n",
      "Epoch 11: 9440 / 10000\n",
      "Epoch 12: 9448 / 10000\n",
      "Epoch 13: 9460 / 10000\n",
      "Epoch 14: 9445 / 10000\n",
      "Epoch 15: 9459 / 10000\n",
      "Epoch 16: 9467 / 10000\n",
      "Epoch 17: 9466 / 10000\n",
      "Epoch 18: 9434 / 10000\n",
      "Epoch 19: 9450 / 10000\n",
      "Epoch 20: 9463 / 10000\n",
      "Epoch 21: 9472 / 10000\n",
      "Epoch 22: 9465 / 10000\n",
      "Epoch 23: 9482 / 10000\n",
      "Epoch 24: 9487 / 10000\n",
      "Epoch 25: 9458 / 10000\n",
      "Epoch 26: 9481 / 10000\n",
      "Epoch 27: 9479 / 10000\n",
      "Epoch 28: 9476 / 10000\n",
      "Epoch 29: 9479 / 10000\n"
     ]
    }
   ],
   "source": [
    "net.SGD(training_data, 30, 10, 3.0, test_data=test_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Challenge 14.3\n",
    "Try creating a network with just two layers - an input and an output layer, no hidden layer - with 784 and 10 neurons, respectively. Train the network using stochastic gradient descent. What classification accuracy can you achieve?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Number of hidden layers\n",
    "\n",
    "Suppose that we want to approximate a set of functions to a given accuracy. How many hidden layers do we need? The answer is: At most two layers, with arbitrary accuracy obtained given enough units per layer. It has been also shown that only one layer is enough to approximate any continuous function. Of course, there is no way to know how many units we would need, and this is not known in general, and this number may grow exponentially with the number of input units. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}