PC_Hopfield.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "Collapsed": "false"
   },
   "source": [
    "# PC-Hopfield Network\n",
    "Purpose: To try training a recurrent PC network as a Hopfield network."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import torch\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "import copy\n",
    "%load_ext autoreload\n",
    "%autoreload 1\n",
    "%aimport PCNetwork\n",
    "%aimport PCLayer\n",
    "%aimport PCConnection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": [
    "dtype = torch.float32\n",
    "if torch.cuda.is_available():\n",
    "    device = torch.device(\"cuda:0\") # Uncomment this to run on GPU\n",
    "else:\n",
    "    device = torch.device(\"cpu\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>.container { width:98% !important; }</style>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from IPython.core.display import display, HTML\n",
    "display(HTML(\"<style>.container { width:98% !important; }</style>\"))\n",
    "np.set_printoptions(edgeitems=30, linewidth=250, formatter=dict(float=lambda x: \"% .4f\" % x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def Hamming(X, Y):\n",
    "    '''\n",
    "     H = Hamming(X, Y)\n",
    "     \n",
    "     Computes the Hamming distance between rows of X and Y.\n",
    "     If X and/or Y have multiple rows, then the output will\n",
    "     be an array.\n",
    "     \n",
    "     Inputs:\n",
    "      X   (M,D) array\n",
    "      Y   (N,D) array\n",
    "      \n",
    "     Output:\n",
    "      H   (M,N) array\n",
    "    '''\n",
    "    H = np.zeros((len(X), len(Y)), dtype=int)\n",
    "    for i,x in enumerate(X):\n",
    "        for j,y in enumerate(Y):\n",
    "            H[i,j] = torch.sum(x*y<=0.)\n",
    "    return H\n",
    "\n",
    "def Perturb(x, p=0., n=0):\n",
    "    '''\n",
    "        y = Perturb(x, p=0., n=0)\n",
    "        \n",
    "        Apply binary noise to x. With probability p, each bit will be randomly\n",
    "        set to -1 or 1.\n",
    "        \n",
    "        Inputs:\n",
    "          x is an array of binary vectors of {-1,1}\n",
    "          p is the probability of each bit being randomly flipped (n=0)\n",
    "          n is the number of bits to flip (p=0)\n",
    "        \n",
    "        Output:\n",
    "          y is an array of binary vectors of {-1,1}\n",
    "    '''\n",
    "    y = copy.deepcopy(x)\n",
    "    for yy in y:\n",
    "        if p!=0.:\n",
    "            for k in range(len(yy)):\n",
    "                if np.random.rand()<p:\n",
    "                    yy[k] = Thresh(np.random.randint(2)*2-1)\n",
    "        if n>0:\n",
    "            k = np.random.randint(len(yy), size=n)\n",
    "            yy[k] *= -1\n",
    "    return y\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "Collapsed": "false"
   },
   "source": [
    "# Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# class SimpleDataset(Dataset):\n",
    "#     def __init__(self, A, B, n=300, noise=0.1):\n",
    "#         self.samples = []\n",
    "#         n_classes = len(A)\n",
    "#         for i in range(n):\n",
    "#             r = np.random.randint(n_classes)\n",
    "#             sample = [A[r]+noise*torch.randn_like(A[r]), B[r]]\n",
    "#             self.samples.append(sample)\n",
    "        \n",
    "#     def __getitem__(self, idx):\n",
    "#         return self.samples[idx]\n",
    "        \n",
    "#     def __len__(self):\n",
    "#         return len(self.samples)\n",
    "\n",
    "# Zeta = torch.rand((4,10,))\n",
    "# print(Zeta)\n",
    "#blah = SimpleDataset(Zeta, Zeta, n=100, noise=0.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class BipolarDataset(Dataset):\n",
    "    def __init__(self, A, n=300, noise=0.1):\n",
    "        self.samples = []\n",
    "        n_classes = len(A)\n",
    "        for i in range(n):\n",
    "            r = np.random.randint(n_classes)\n",
    "            blah = Perturb(A[r], p=0.)\n",
    "            sample = [blah, blah]\n",
    "            self.samples.append(sample)\n",
    "        \n",
    "    def __getitem__(self, idx):\n",
    "        return self.samples[idx]\n",
    "        \n",
    "    def __len__(self):\n",
    "        return len(self.samples)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12 target patterns\n",
      "Hamming Distances:\n",
      "[[0 4 2 3]\n",
      " [4 0 6 7]\n",
      " [2 6 0 3]\n",
      " [3 7 3 0]]\n"
     ]
    }
   ],
   "source": [
    "# Random binary Vectors\n",
    "X = np.random.binomial(n=1, p=0.5, size=(4,12))\n",
    "Zeta = torch.tensor(2*X - 1, dtype=torch.float)\n",
    "Zeta_c = 10.*Zeta\n",
    "N = np.shape(Zeta)[1]\n",
    "print(f'{N} target patterns')\n",
    "print(f'Hamming Distances:\\n{np.array(Hamming(Zeta, Zeta), dtype=int)}')\n",
    "\n",
    "bipolar_ds = BipolarDataset(Zeta_c, n=30, noise=0.)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "Collapsed": "false"
   },
   "source": [
    "# Create Network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": [
    "N_neurons = 12\n",
    "net = PCNetwork.PCNetwork()\n",
    "\n",
    "net.AddLayer(PCLayer.PCLayer(n=N_neurons))  # 0 (v0)\n",
    "net.AddLayer(PCLayer.PCLayer(n=N_neurons, type='error'))  # 1 (e1)\n",
    "net.AddLayer(PCLayer.PCLayer(n=N_neurons))  # 2 (v1)\n",
    "\n",
    "net.Connect(v_idx=0, e_idx=1, type='1to1')\n",
    "net.Connect(e_idx=1, v_idx=2, type='1to1')\n",
    "net.Connect(v_idx=2, e_idx=1, lower_layer=2, type='general', sym=True, act_text='tanh')\n",
    "\n",
    "net.SetTau(0.05)\n",
    "net.SetGamma(0.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/jorchard/Dropbox/research/pc_network/PCConnection.py:379: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  self.M = torch.tensor(deepcopy(M))\n"
     ]
    }
   ],
   "source": [
    "net.con[0].SetIdentity(mult=0.)\n",
    "net.con[2].SetRandom(random=0.7)\n",
    "XX = Zeta_c.transpose(1,0)@Zeta_c / N_neurons / 4.\n",
    "XX -= torch.diag(XX)\n",
    "net.con[2].SetWeights(XX.clone().detach())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(1.)\n",
      "tensor([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]])\n",
      "tensor(-1.)\n",
      "tensor([[1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0.],\n",
      "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1.]])\n",
      "tensor(1.)\n",
      "tensor([[  0.0000,  -4.1667,  -4.1667, -12.5000,  -4.1667,   0.0000,  -4.1667,\n",
      "          -4.1667, -12.5000, -16.6667, -16.6667,  -8.3333],\n",
      "        [ -4.1667,   0.0000,  -8.3333, -16.6667,  -8.3333,  -4.1667,  -8.3333,\n",
      "           0.0000, -16.6667, -12.5000, -12.5000, -12.5000],\n",
      "        [ -4.1667,  -8.3333,   0.0000,  -8.3333,  -8.3333,  -4.1667,   0.0000,\n",
      "          -8.3333,  -8.3333, -12.5000, -12.5000,  -4.1667],\n",
      "        [-12.5000, -16.6667,  -8.3333,   0.0000,  -8.3333, -12.5000,  -8.3333,\n",
      "         -16.6667,   0.0000,  -4.1667,  -4.1667,  -4.1667],\n",
      "        [ -4.1667,  -8.3333,  -8.3333,  -8.3333,   0.0000,  -4.1667,  -8.3333,\n",
      "          -8.3333,  -8.3333, -12.5000, -12.5000,  -4.1667],\n",
      "        [  0.0000,  -4.1667,  -4.1667, -12.5000,  -4.1667,   0.0000,  -4.1667,\n",
      "          -4.1667, -12.5000, -16.6667, -16.6667,  -8.3333],\n",
      "        [ -4.1667,  -8.3333,   0.0000,  -8.3333,  -8.3333,  -4.1667,   0.0000,\n",
      "          -8.3333,  -8.3333, -12.5000, -12.5000,  -4.1667],\n",
      "        [ -4.1667,   0.0000,  -8.3333, -16.6667,  -8.3333,  -4.1667,  -8.3333,\n",
      "           0.0000, -16.6667, -12.5000, -12.5000, -12.5000],\n",
      "        [-12.5000, -16.6667,  -8.3333,   0.0000,  -8.3333, -12.5000,  -8.3333,\n",
      "         -16.6667,   0.0000,  -4.1667,  -4.1667,  -4.1667],\n",
      "        [-16.6667, -12.5000, -12.5000,  -4.1667, -12.5000, -16.6667, -12.5000,\n",
      "         -12.5000,  -4.1667,   0.0000,   0.0000,  -8.3333],\n",
      "        [-16.6667, -12.5000, -12.5000,  -4.1667, -12.5000, -16.6667, -12.5000,\n",
      "         -12.5000,  -4.1667,   0.0000,   0.0000,  -8.3333],\n",
      "        [ -8.3333, -12.5000,  -4.1667,  -4.1667,  -4.1667,  -8.3333,  -4.1667,\n",
      "         -12.5000,  -4.1667,  -8.3333,  -8.3333,   0.0000]])\n"
     ]
    }
   ],
   "source": [
    "for c in net.con:\n",
    "    print(c.M_sign)\n",
    "    print(c.M)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])\n",
      "tensor([0., 0., 0., 0., -0., 0., 0., 0., 0., 0., 0., 0.])\n",
      "tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])\n"
     ]
    }
   ],
   "source": [
    "for l in net.lyr:\n",
    "    print(l.bias)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": [
    "#net.Save('my.net')\n",
    "#net2 = PCNetwork.PCNetwork.Load('my.net')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "Collapsed": "false"
   },
   "source": [
    "# Train it"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "bipolar_ds = BipolarDataset(Zeta_c, n=10, noise=0.)\n",
    "bipolar_dl = DataLoader(bipolar_ds, batch_size=4, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[tensor([-10., -10.,  10.,  10., -10., -10.,  10., -10.,  10.,  10.,  10.,  10.]),\n",
       "  tensor([-10., -10.,  10.,  10., -10., -10.,  10., -10.,  10.,  10.,  10.,  10.])],\n",
       " [tensor([-10., -10., -10.,  10.,  10., -10., -10., -10.,  10.,  10.,  10.,  10.]),\n",
       "  tensor([-10., -10., -10.,  10.,  10., -10., -10., -10.,  10.,  10.,  10.,  10.])],\n",
       " [tensor([-10.,  10., -10., -10., -10., -10., -10.,  10., -10.,  10.,  10., -10.]),\n",
       "  tensor([-10.,  10., -10., -10., -10., -10., -10.,  10., -10.,  10.,  10., -10.])],\n",
       " [tensor([-10.,  10., -10., -10., -10., -10., -10.,  10., -10.,  10.,  10., -10.]),\n",
       "  tensor([-10.,  10., -10., -10., -10., -10., -10.,  10., -10.,  10.,  10., -10.])],\n",
       " [tensor([-10., -10., -10.,  10.,  10., -10., -10., -10.,  10.,  10.,  10.,  10.]),\n",
       "  tensor([-10., -10., -10.,  10.,  10., -10., -10., -10.,  10.,  10.,  10.,  10.])],\n",
       " [tensor([-10., -10.,  10.,  10., -10., -10.,  10., -10.,  10.,  10.,  10.,  10.]),\n",
       "  tensor([-10., -10.,  10.,  10., -10., -10.,  10., -10.,  10.,  10.,  10.,  10.])],\n",
       " [tensor([-10., -10., -10.,  10., -10., -10., -10., -10.,  10.,  10.,  10., -10.]),\n",
       "  tensor([-10., -10., -10.,  10., -10., -10., -10., -10.,  10.,  10.,  10., -10.])],\n",
       " [tensor([-10., -10., -10.,  10.,  10., -10., -10., -10.,  10.,  10.,  10.,  10.]),\n",
       "  tensor([-10., -10., -10.,  10.,  10., -10., -10., -10.,  10.,  10.,  10.,  10.])],\n",
       " [tensor([-10.,  10., -10., -10., -10., -10., -10.,  10., -10.,  10.,  10., -10.]),\n",
       "  tensor([-10.,  10., -10., -10., -10., -10., -10.,  10., -10.,  10.,  10., -10.])],\n",
       " [tensor([-10., -10., -10.,  10.,  10., -10., -10., -10.,  10.,  10.,  10.,  10.]),\n",
       "  tensor([-10., -10., -10.,  10.,  10., -10., -10., -10.,  10.,  10.,  10.,  10.])]]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bipolar_ds.samples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 10/10 [00:07<00:00,  1.28it/s]\n"
     ]
    }
   ],
   "source": [
    "net.Reset()\n",
    "net.Learn(bipolar_dl, T=1., epochs=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([-4.3214, -2.7751, -2.4092,  4.0084, -2.0342, -4.3136, -2.4051, -2.7765,\n",
       "         4.0108,  5.5565,  5.5578, -0.1196])"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.lyr[1].bias"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ -4.3207,  -6.9450,  -6.5738,  -8.4866,  -6.2043,  -4.3226,  -6.5745,\n",
       "          -6.9449,  -8.4871, -11.1112, -11.1116,  -8.4577],\n",
       "        [ -8.3335,  -5.8333,  -8.3333, -10.8335,  -8.3334,  -8.3334,  -8.3334,\n",
       "          -5.8334, -10.8335,  -8.3333,  -8.3333,  -8.3333],\n",
       "        [ -0.8642,  -1.3890,   2.6852,  -1.6973,  -3.2408,  -0.8645,   2.6851,\n",
       "          -1.3889,  -1.6974,  -2.2223,  -2.2223,   0.3084],\n",
       "        [ -8.3332, -10.8333,  -8.3334,  -5.8332,  -8.3333,  -8.3333,  -8.3334,\n",
       "         -10.8335,  -5.8332,  -8.3333,  -8.3333,  -8.3334],\n",
       "        [ -1.7283,  -2.7780,  -4.6295,  -3.3946,   3.5183,  -1.7291,  -4.6298,\n",
       "          -2.7779,  -3.3948,  -4.4446,  -4.4446,   0.6169],\n",
       "        [ -4.3207,  -6.9450,  -6.5738,  -8.4866,  -6.2043,  -4.3226,  -6.5745,\n",
       "          -6.9449,  -8.4871, -11.1112, -11.1116,  -8.4577],\n",
       "        [ -0.8642,  -1.3890,   2.6852,  -1.6973,  -3.2408,  -0.8645,   2.6851,\n",
       "          -1.3889,  -1.6974,  -2.2223,  -2.2223,   0.3084],\n",
       "        [ -8.3335,  -5.8333,  -8.3333, -10.8335,  -8.3334,  -8.3334,  -8.3334,\n",
       "          -5.8334, -10.8335,  -8.3333,  -8.3333,  -8.3333],\n",
       "        [ -8.3332, -10.8333,  -8.3334,  -5.8332,  -8.3333,  -8.3333,  -8.3334,\n",
       "         -10.8335,  -5.8332,  -8.3333,  -8.3333,  -8.3334],\n",
       "        [-12.3458,  -9.7215, -10.0929,  -8.1800, -10.4624, -12.3440, -10.0920,\n",
       "          -9.7219,  -8.1795,  -5.5553,  -5.5551,  -8.2091],\n",
       "        [-12.3458,  -9.7215, -10.0929,  -8.1800, -10.4624, -12.3440, -10.0920,\n",
       "          -9.7219,  -8.1795,  -5.5553,  -5.5551,  -8.2091],\n",
       "        [  1.7283,   2.7780,   4.6295,   3.3946,   6.4817,   1.7291,   4.6299,\n",
       "           2.7779,   3.3947,   4.4446,   4.4446,   9.3831]])"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.con[2].M"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test it"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "net.Reset()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "foo = Zeta_c.clone().detach() / 1.\n",
    "net.SetInput(foo)\n",
    "net.lyr[-1].SetState(foo, random=0.01)\n",
    "foo = Perturb(foo, n=2)\n",
    "net.lyr[-1].SetState(foo)\n",
    "#net.lyr[-1].ResetState(random=0.1)\n",
    "net.lyr[0].Clamped(True)\n",
    "net.lyr[-1].Clamped(False)\n",
    "net.lyr[1].ResetState(random=0.1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[2, 2, 4, 5],\n",
       "       [4, 2, 6, 7],\n",
       "       [4, 6, 2, 3],\n",
       "       [3, 5, 3, 2]])"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Hamming(foo, Zeta_c)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-10., -10., -10., -10., -10., -10., -10.,  10.,  10.,  10.,  10., -10.],\n",
       "        [ 10.,  10., -10., -10., -10., -10., -10.,  10.,  10.,  10.,  10., -10.],\n",
       "        [-10.,  10., -10.,  10.,  10., -10.,  10., -10.,  10.,  10.,  10.,  10.],\n",
       "        [-10., -10., -10., -10., -10., -10.,  10., -10.,  10.,  10.,  10.,  10.]])"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.lyr[-1].x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": [
    "net.lyr[0].Probe(True)\n",
    "net.lyr[1].Probe(True)\n",
    "net.lyr[2].Probe(True)\n",
    "net.Learning(False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "net.Run(1., dt=0.001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false",
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "net.Plot(idx=0);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1174,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 2,  5, 12,  5],\n",
       "       [ 5,  0,  7,  6],\n",
       "       [ 7,  6,  3,  6],\n",
       "       [ 7,  6,  7,  0]])"
      ]
     },
     "execution_count": 1174,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Hamming(Zeta_c, net.lyr[-1].x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1129,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-29.1667,  37.5000, -12.5000, -29.1667, -37.5000,  45.8333, -45.8333,\n",
       "         -45.8333,  37.5000,  12.5000,  20.8333,  37.5000],\n",
       "        [ -5.1658,  28.7621,  -4.9907,  -5.1658, -29.7463,  37.3407, -38.3248,\n",
       "         -38.3248,  12.7602,  20.0085,  28.5871,  12.7602],\n",
       "        [ 29.1667, -37.5000,  12.5000,  29.1667,  37.5000, -45.8333,  45.8333,\n",
       "          45.8333, -37.5000, -12.5000, -20.8333, -37.5000],\n",
       "        [  4.1686, -37.4801, -45.8194,   4.1686,   4.1707,  12.5139, -45.8234,\n",
       "         -45.8234,  12.5159, -37.4861,  12.5080,  12.5159]])"
      ]
     },
     "execution_count": 1129,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.con[2].sigma() @ net.con[2].M"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1130,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-29.1666,  37.5000, -12.5000, -29.1666, -37.5000,  45.8333, -45.8334,\n",
       "         -45.8334,  37.5000,  12.5000,  20.8334,  37.5000],\n",
       "        [ -5.1412,  28.7621,  -4.9680,   2.2734, -29.7463,  37.3407, -38.3249,\n",
       "         -38.3249,  -1.5899,  20.0085,  28.5871,  12.7601],\n",
       "        [ 29.1666, -37.5000,  12.5000,  29.1666,  37.5000, -45.8333,  45.8334,\n",
       "          45.8334, -37.5000, -12.5000, -20.8334, -37.5000],\n",
       "        [  4.1687, -37.4800, -45.8193,   4.1687,   4.1707,  12.5139, -45.8233,\n",
       "         -45.8233,  12.5159, -37.4860,  12.5080,  12.5159]])"
      ]
     },
     "execution_count": 1130,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.lyr[-1].x"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Many Perturbed Runs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 922,
   "metadata": {},
   "outputs": [],
   "source": [
    "x0 = Zeta_c[1]\n",
    "ins = torch.vstack((x0, x0, x0, x0, x0, x0))\n",
    "ins += torch.normal(torch.zeros_like(ins))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 923,
   "metadata": {},
   "outputs": [],
   "source": [
    "net.Reset()\n",
    "net.SetInput(ins)\n",
    "net.lyr[-1].SetState(ins)\n",
    "#net.lyr[-1].ResetState(random=0.1)\n",
    "net.lyr[0].Clamped(True)\n",
    "net.lyr[-1].Clamped(False)\n",
    "net.lyr[1].ResetState(random=0.1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 924,
   "metadata": {},
   "outputs": [],
   "source": [
    "net.Run(1., dt=0.001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 925,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-55000.9922, -45000.7852,   5000.0864, -55000.9922,  15000.2666,\n",
       "          15000.2666,  25000.4453,  25000.4453,  15000.2666, -45000.7852,\n",
       "           5000.0864, -45000.7852],\n",
       "        [-55000.9922, -45000.7852,   5000.0864, -55000.9922,  15000.2666,\n",
       "          15000.2666,  25000.4453,  25000.4473,  15000.2666, -45000.7852,\n",
       "           5000.0864, -45000.7852],\n",
       "        [-55000.9922, -45000.7852,   5000.0864, -55000.9922,  15000.2666,\n",
       "          15000.2666,  25000.4473,  25000.4473,  15000.2666, -45000.7852,\n",
       "           5000.0864, -45000.7852],\n",
       "        [-55000.9922, -45000.7852,   5000.0864, -55000.9922,  15000.2666,\n",
       "          15000.2666,  25000.4473,  25000.4473,  15000.2666, -45000.7852,\n",
       "           5000.0864, -45000.7852],\n",
       "        [-55000.9922, -45000.7852,   5000.0864, -55000.9922,  15000.2666,\n",
       "          15000.2666,  25000.4453,  25000.4473,  15000.2666, -45000.7852,\n",
       "           5000.0864, -45000.7852],\n",
       "        [-55000.9922, -45000.7852,   5000.0864, -55000.9922,  15000.2666,\n",
       "          15000.2666,  25000.4473,  25000.4473,  15000.2666, -45000.7852,\n",
       "           5000.0864, -45000.7852]])"
      ]
     },
     "execution_count": 925,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.lyr[-1].x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 926,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-55000., -45000.,   5000., -55000.,  15000.,  15000.,  25000.,  25000.,\n",
       "          15000., -45000.,   5000., -45000.],\n",
       "        [-55000., -45000.,   5000., -55000.,  15000.,  15000.,  25000.,  25000.,\n",
       "          15000., -45000.,   5000., -45000.],\n",
       "        [-55000., -45000.,   5000., -55000.,  15000.,  15000.,  25000.,  25000.,\n",
       "          15000., -45000.,   5000., -45000.],\n",
       "        [-55000., -45000.,   5000., -55000.,  15000.,  15000.,  25000.,  25000.,\n",
       "          15000., -45000.,   5000., -45000.],\n",
       "        [-55000., -45000.,   5000., -55000.,  15000.,  15000.,  25000.,  25000.,\n",
       "          15000., -45000.,   5000., -45000.],\n",
       "        [-55000., -45000.,   5000., -55000.,  15000.,  15000.,  25000.,  25000.,\n",
       "          15000., -45000.,   5000., -45000.]])"
      ]
     },
     "execution_count": 926,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.con[2].sigma() @ net.con[2].M"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 927,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "net.Plot();"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.3"
  },
  "toc-showcode": false,
  "toc-showtags": false
 },
 "nbformat": 4,
 "nbformat_minor": 4
}