Examples.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Introduction\n",
    "As described in the repo README, the experiment runner provides the fixture for running experiments on UCI data sets. One can also use this repo as a module providing an API to the RPA-GP family of GP models. \n",
    "\n",
    "Below are demonstrations on how to create and use these models on example data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7.4.0\n"
     ]
    }
   ],
   "source": [
    "# Preliminary imports\n",
    "import torch\n",
    "from torch.optim import Adam\n",
    "from gpytorch.kernels import RBFKernel, ScaleKernel, AdditiveStructureKernel\n",
    "from gpytorch.likelihoods import GaussianLikelihood\n",
    "from gpytorch.mlls import ExactMarginalLogLikelihood"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gp_experiment_runner import _determine_folds, _access_fold, _normalize_by_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load a data set\n",
    "from gp_experiment_runner import load_dataset  # assumes UCI datasets are available locally\n",
    "dataset = load_dataset('yacht') # returns a Pandas dataframe\n",
    "\n",
    "n_train = len(dataset) // 10 * 9\n",
    "train = dataset.iloc[:n_train]\n",
    "test = dataset.iloc[n_train:]\n",
    "\n",
    "# Or use utilities from repo:\n",
    "# from gp_experiment_runner import _determine_folds, _access_fold, _normalize_by_train\n",
    "# folds = _determine_folds(0.1, dataset)\n",
    "# train, test = _access_fold(dataset, folds, 0)\n",
    "# train, test = _normalize_by_train(train, test)\n",
    "\n",
    "train_x = torch.tensor(train.iloc[:, :-1].values, dtype=torch.float)\n",
    "train_y = torch.tensor(train.iloc[:, -1].values, dtype=torch.float)\n",
    "test_x = torch.tensor(test.iloc[:, :-1].values, dtype=torch.float)\n",
    "test_y = torch.tensor(test.iloc[:, -1].values, dtype=torch.float)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create an RPA-GP model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gp_models import ExactGPModel, ScaledProjectionKernel\n",
    "import rp\n",
    "n, d = train_x.shape\n",
    "num_projs = 20\n",
    "\n",
    "# Draw random projections and store in a linear module\n",
    "# Here, we are drawing 20 Gaussian projections into 1 dimension.\n",
    "projs = [rp.gen_rp(d, 1, dist='gaussian') for _ in range(num_projs)]\n",
    "proj_module = torch.nn.Linear(d, num_projs, bias=False)\n",
    "proj_module.weight.data = torch.cat(projs, dim=1).t()\n",
    "\n",
    "# Create the additive model that operates over these projections\n",
    "# Fixing the outputscale and lengthscale of the base kernels.\n",
    "base_kernel = RBFKernel()\n",
    "base_kernel.initialize(lengthscale=torch.tensor([1.]))\n",
    "base_kernel = ScaleKernel(base_kernel)\n",
    "base_kernel.initialize(outputscale=torch.tensor([1/num_projs]))\n",
    "\n",
    "# Combine into a single module.\n",
    "kernel = ScaledProjectionKernel(proj_module, base_kernel, \n",
    "                                prescale=False,\n",
    "                                ard_num_dims=num_projs,\n",
    "                                learn_proj=False)\n",
    "# Or, just call the method from training_routines that wraps this initialization\n",
    "# from training_routines import create_additive_rp_kernel\n",
    "# create_additive_rp_kernel(d, num_projs, learn_proj=False, kernel_type='RBF', \n",
    "#                           space_proj=False, prescale=False, ard=True, k=1, \n",
    "#                           proj_dist='gaussian')\n",
    "\n",
    "kernel = ScaleKernel(kernel) # Optionally wrap with an additional ScaleKernel \n",
    "\n",
    "# Create an ExactGP model with this kernel\n",
    "likelihood = GaussianLikelihood()\n",
    "likelihood.noise = 1.\n",
    "model = ExactGPModel(train_x, train_y, likelihood, kernel)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 0: Loss=1.4288, Noise=0.9937\n",
      "Iter 1: Loss=1.4287, Noise=0.9893\n",
      "Iter 2: Loss=1.4288, Noise=0.9862\n",
      "Iter 3: Loss=1.4290, Noise=0.9824\n",
      "Iter 4: Loss=1.4288, Noise=0.9824\n",
      "Iter 5: Loss=1.4289, Noise=0.9836\n",
      "Iter 6: Loss=1.4286, Noise=0.9857\n",
      "Iter 7: Loss=1.4286, Noise=0.9879\n",
      "Iter 8: Loss=1.4285, Noise=0.9907\n",
      "Iter 9: Loss=1.4286, Noise=0.9941\n",
      "Iter 10: Loss=1.4288, Noise=0.9972\n",
      "Iter 11: Loss=1.4286, Noise=0.9992\n",
      "Iter 12: Loss=1.4287, Noise=0.9998\n",
      "Iter 13: Loss=1.4288, Noise=0.9990\n",
      "Iter 14: Loss=1.4287, Noise=0.9976\n",
      "Iter 15: Loss=1.4287, Noise=0.9952\n",
      "Iter 16: Loss=1.4288, Noise=0.9918\n",
      "Iter 17: Loss=1.4288, Noise=0.9883\n",
      "Iter 18: Loss=1.4288, Noise=0.9851\n",
      "Iter 19: Loss=1.4287, Noise=0.9811\n",
      "Iter 20: Loss=1.4286, Noise=0.9770\n",
      "Iter 21: Loss=1.4289, Noise=0.9738\n",
      "Iter 22: Loss=1.4286, Noise=0.9699\n",
      "Iter 23: Loss=1.4287, Noise=0.9671\n",
      "Iter 24: Loss=1.4288, Noise=0.9654\n",
      "Iter 25: Loss=1.4288, Noise=0.9635\n",
      "Iter 26: Loss=1.4289, Noise=0.9611\n",
      "Iter 27: Loss=1.4289, Noise=0.9600\n",
      "Iter 28: Loss=1.4289, Noise=0.9592\n",
      "Iter 29: Loss=1.4288, Noise=0.9593\n",
      "Iter 30: Loss=1.4288, Noise=0.9602\n",
      "Iter 31: Loss=1.4287, Noise=0.9617\n",
      "Iter 32: Loss=1.4287, Noise=0.9636\n",
      "Iter 33: Loss=1.4290, Noise=0.9652\n",
      "Iter 34: Loss=1.4288, Noise=0.9674\n",
      "Iter 35: Loss=1.4286, Noise=0.9697\n",
      "Iter 36: Loss=1.4287, Noise=0.9730\n",
      "Iter 37: Loss=1.4288, Noise=0.9771\n",
      "Iter 38: Loss=1.4288, Noise=0.9818\n",
      "Iter 39: Loss=1.4287, Noise=0.9862\n",
      "Iter 40: Loss=1.4285, Noise=0.9897\n",
      "Iter 41: Loss=1.4287, Noise=0.9940\n",
      "Iter 42: Loss=1.4288, Noise=0.9976\n",
      "Iter 43: Loss=1.4287, Noise=1.0002\n",
      "Iter 44: Loss=1.4287, Noise=1.0009\n",
      "Iter 45: Loss=1.4287, Noise=1.0006\n",
      "Iter 46: Loss=1.4288, Noise=0.9999\n",
      "Iter 47: Loss=1.4285, Noise=0.9990\n",
      "Iter 48: Loss=1.4288, Noise=0.9976\n",
      "Iter 49: Loss=1.4288, Noise=0.9966\n",
      "Iter 50: Loss=1.4288, Noise=0.9957\n",
      "Iter 51: Loss=1.4286, Noise=0.9949\n",
      "Iter 52: Loss=1.4287, Noise=0.9937\n",
      "Iter 53: Loss=1.4287, Noise=0.9932\n",
      "Iter 54: Loss=1.4287, Noise=0.9931\n",
      "Iter 55: Loss=1.4289, Noise=0.9931\n",
      "Iter 56: Loss=1.4287, Noise=0.9934\n",
      "Iter 57: Loss=1.4289, Noise=0.9935\n",
      "Iter 58: Loss=1.4288, Noise=0.9935\n",
      "Iter 59: Loss=1.4287, Noise=0.9928\n",
      "Iter 60: Loss=1.4287, Noise=0.9923\n",
      "Iter 61: Loss=1.4285, Noise=0.9914\n",
      "Iter 62: Loss=1.4288, Noise=0.9903\n",
      "Iter 63: Loss=1.4285, Noise=0.9882\n",
      "Iter 64: Loss=1.4286, Noise=0.9869\n",
      "Iter 65: Loss=1.4286, Noise=0.9859\n",
      "Iter 66: Loss=1.4288, Noise=0.9846\n",
      "Iter 67: Loss=1.4288, Noise=0.9843\n",
      "Iter 68: Loss=1.4287, Noise=0.9841\n",
      "Iter 69: Loss=1.4287, Noise=0.9837\n",
      "Iter 70: Loss=1.4284, Noise=0.9839\n",
      "Iter 71: Loss=1.4288, Noise=0.9838\n",
      "Iter 72: Loss=1.4286, Noise=0.9843\n",
      "Iter 73: Loss=1.4286, Noise=0.9844\n",
      "Iter 74: Loss=1.4286, Noise=0.9850\n",
      "Iter 75: Loss=1.4286, Noise=0.9858\n",
      "Iter 76: Loss=1.4287, Noise=0.9864\n",
      "Iter 77: Loss=1.4286, Noise=0.9877\n",
      "Iter 78: Loss=1.4287, Noise=0.9892\n",
      "Iter 79: Loss=1.4287, Noise=0.9906\n",
      "Iter 80: Loss=1.4288, Noise=0.9920\n",
      "Iter 81: Loss=1.4286, Noise=0.9930\n",
      "Iter 82: Loss=1.4285, Noise=0.9937\n",
      "Iter 83: Loss=1.4285, Noise=0.9942\n",
      "Iter 84: Loss=1.4286, Noise=0.9948\n",
      "Iter 85: Loss=1.4288, Noise=0.9958\n",
      "Iter 86: Loss=1.4289, Noise=0.9964\n",
      "Iter 87: Loss=1.4288, Noise=0.9973\n",
      "Iter 88: Loss=1.4288, Noise=0.9974\n",
      "Iter 89: Loss=1.4287, Noise=0.9974\n",
      "Iter 90: Loss=1.4287, Noise=0.9971\n",
      "Iter 91: Loss=1.4288, Noise=0.9970\n",
      "Iter 92: Loss=1.4287, Noise=0.9971\n",
      "Iter 93: Loss=1.4289, Noise=0.9977\n",
      "Iter 94: Loss=1.4286, Noise=0.9976\n",
      "Iter 95: Loss=1.4288, Noise=0.9972\n",
      "Iter 96: Loss=1.4287, Noise=0.9973\n",
      "Iter 97: Loss=1.4287, Noise=0.9971\n",
      "Iter 98: Loss=1.4287, Noise=0.9965\n",
      "Iter 99: Loss=1.4286, Noise=0.9965\n",
      "Iter 100: Loss=1.4287, Noise=0.9967\n",
      "Iter 101: Loss=1.4288, Noise=0.9973\n",
      "Iter 102: Loss=1.4286, Noise=0.9974\n",
      "Iter 103: Loss=1.4288, Noise=0.9981\n",
      "Iter 104: Loss=1.4287, Noise=0.9986\n",
      "Iter 105: Loss=1.4288, Noise=0.9993\n",
      "Iter 106: Loss=1.4286, Noise=0.9993\n",
      "Iter 107: Loss=1.4288, Noise=0.9985\n",
      "Iter 108: Loss=1.4288, Noise=0.9981\n",
      "Iter 109: Loss=1.4289, Noise=0.9976\n",
      "Iter 110: Loss=1.4289, Noise=0.9973\n",
      "Iter 111: Loss=1.4286, Noise=0.9962\n",
      "Iter 112: Loss=1.4286, Noise=0.9947\n",
      "Iter 113: Loss=1.4288, Noise=0.9939\n",
      "Iter 114: Loss=1.4286, Noise=0.9943\n",
      "Iter 115: Loss=1.4287, Noise=0.9941\n",
      "Iter 116: Loss=1.4286, Noise=0.9931\n",
      "Iter 117: Loss=1.4287, Noise=0.9925\n",
      "Iter 118: Loss=1.4287, Noise=0.9916\n",
      "Iter 119: Loss=1.4288, Noise=0.9903\n",
      "Iter 120: Loss=1.4287, Noise=0.9881\n",
      "Iter 121: Loss=1.4285, Noise=0.9859\n",
      "Iter 122: Loss=1.4285, Noise=0.9839\n",
      "Iter 123: Loss=1.4287, Noise=0.9824\n",
      "Iter 124: Loss=1.4289, Noise=0.9812\n",
      "Iter 125: Loss=1.4286, Noise=0.9804\n",
      "Iter 126: Loss=1.4287, Noise=0.9802\n",
      "Iter 127: Loss=1.4289, Noise=0.9800\n",
      "Iter 128: Loss=1.4287, Noise=0.9800\n",
      "Iter 129: Loss=1.4287, Noise=0.9800\n",
      "Iter 130: Loss=1.4287, Noise=0.9789\n",
      "Iter 131: Loss=1.4288, Noise=0.9770\n",
      "Iter 132: Loss=1.4287, Noise=0.9766\n",
      "Iter 133: Loss=1.4287, Noise=0.9766\n",
      "Iter 134: Loss=1.4285, Noise=0.9766\n",
      "Iter 135: Loss=1.4287, Noise=0.9773\n",
      "Iter 136: Loss=1.4286, Noise=0.9778\n",
      "Iter 137: Loss=1.4285, Noise=0.9791\n",
      "Iter 138: Loss=1.4286, Noise=0.9794\n",
      "Iter 139: Loss=1.4286, Noise=0.9790\n",
      "Iter 140: Loss=1.4287, Noise=0.9782\n",
      "Iter 141: Loss=1.4287, Noise=0.9781\n",
      "Iter 142: Loss=1.4287, Noise=0.9789\n",
      "Iter 143: Loss=1.4287, Noise=0.9791\n",
      "Iter 144: Loss=1.4286, Noise=0.9794\n",
      "Iter 145: Loss=1.4286, Noise=0.9797\n",
      "Iter 146: Loss=1.4288, Noise=0.9798\n",
      "Iter 147: Loss=1.4289, Noise=0.9796\n",
      "Iter 148: Loss=1.4286, Noise=0.9804\n",
      "Iter 149: Loss=1.4287, Noise=0.9820\n",
      "Iter 150: Loss=1.4287, Noise=0.9833\n",
      "Iter 151: Loss=1.4287, Noise=0.9851\n",
      "Iter 152: Loss=1.4288, Noise=0.9860\n",
      "Iter 153: Loss=1.4288, Noise=0.9871\n",
      "Iter 154: Loss=1.4286, Noise=0.9877\n",
      "Iter 155: Loss=1.4286, Noise=0.9880\n",
      "Iter 156: Loss=1.4286, Noise=0.9870\n",
      "Iter 157: Loss=1.4286, Noise=0.9848\n",
      "Iter 158: Loss=1.4287, Noise=0.9829\n",
      "Iter 159: Loss=1.4286, Noise=0.9811\n",
      "Iter 160: Loss=1.4287, Noise=0.9791\n",
      "Iter 161: Loss=1.4287, Noise=0.9779\n",
      "Iter 162: Loss=1.4289, Noise=0.9767\n",
      "Iter 163: Loss=1.4289, Noise=0.9757\n",
      "Iter 164: Loss=1.4289, Noise=0.9742\n",
      "Iter 165: Loss=1.4287, Noise=0.9739\n",
      "Iter 166: Loss=1.4288, Noise=0.9740\n",
      "Iter 167: Loss=1.4287, Noise=0.9747\n",
      "Iter 168: Loss=1.4286, Noise=0.9745\n",
      "Iter 169: Loss=1.4288, Noise=0.9745\n",
      "Iter 170: Loss=1.4288, Noise=0.9744\n",
      "Iter 171: Loss=1.4286, Noise=0.9754\n",
      "Iter 172: Loss=1.4289, Noise=0.9759\n",
      "Iter 173: Loss=1.4287, Noise=0.9764\n",
      "Iter 174: Loss=1.4289, Noise=0.9762\n",
      "Iter 175: Loss=1.4286, Noise=0.9761\n",
      "Iter 176: Loss=1.4288, Noise=0.9756\n",
      "Iter 177: Loss=1.4287, Noise=0.9752\n",
      "Iter 178: Loss=1.4286, Noise=0.9751\n",
      "Iter 179: Loss=1.4287, Noise=0.9755\n",
      "Iter 180: Loss=1.4288, Noise=0.9757\n",
      "Iter 181: Loss=1.4288, Noise=0.9759\n",
      "Iter 182: Loss=1.4288, Noise=0.9764\n",
      "Iter 183: Loss=1.4286, Noise=0.9763\n",
      "Iter 184: Loss=1.4286, Noise=0.9760\n",
      "Iter 185: Loss=1.4288, Noise=0.9761\n",
      "Iter 186: Loss=1.4288, Noise=0.9767\n",
      "Iter 187: Loss=1.4289, Noise=0.9778\n",
      "Iter 188: Loss=1.4286, Noise=0.9783\n",
      "Iter 189: Loss=1.4285, Noise=0.9784\n",
      "Iter 190: Loss=1.4286, Noise=0.9787\n",
      "Iter 191: Loss=1.4286, Noise=0.9789\n",
      "Iter 192: Loss=1.4288, Noise=0.9798\n",
      "Iter 193: Loss=1.4286, Noise=0.9814\n",
      "Iter 194: Loss=1.4286, Noise=0.9822\n",
      "Iter 195: Loss=1.4287, Noise=0.9827\n",
      "Iter 196: Loss=1.4286, Noise=0.9834\n",
      "Iter 197: Loss=1.4286, Noise=0.9852\n",
      "Iter 198: Loss=1.4287, Noise=0.9869\n",
      "Iter 199: Loss=1.4287, Noise=0.9886\n",
      "Iter 200: Loss=1.4288, Noise=0.9899\n",
      "Iter 201: Loss=1.4287, Noise=0.9908\n",
      "Iter 202: Loss=1.4286, Noise=0.9906\n",
      "Iter 203: Loss=1.4286, Noise=0.9906\n",
      "Iter 204: Loss=1.4286, Noise=0.9895\n",
      "Iter 205: Loss=1.4289, Noise=0.9892\n",
      "Iter 206: Loss=1.4286, Noise=0.9891\n",
      "Iter 207: Loss=1.4288, Noise=0.9887\n",
      "Iter 208: Loss=1.4289, Noise=0.9880\n",
      "Iter 209: Loss=1.4286, Noise=0.9879\n",
      "Iter 210: Loss=1.4289, Noise=0.9877\n",
      "Iter 211: Loss=1.4286, Noise=0.9874\n",
      "Iter 212: Loss=1.4287, Noise=0.9874\n",
      "Iter 213: Loss=1.4287, Noise=0.9876\n",
      "Iter 214: Loss=1.4286, Noise=0.9873\n",
      "Iter 215: Loss=1.4287, Noise=0.9865\n",
      "Iter 216: Loss=1.4285, Noise=0.9855\n",
      "Iter 217: Loss=1.4288, Noise=0.9846\n",
      "Iter 218: Loss=1.4288, Noise=0.9837\n",
      "Iter 219: Loss=1.4285, Noise=0.9821\n",
      "Iter 220: Loss=1.4288, Noise=0.9826\n",
      "Iter 221: Loss=1.4284, Noise=0.9834\n",
      "Iter 222: Loss=1.4286, Noise=0.9837\n",
      "Iter 223: Loss=1.4288, Noise=0.9846\n",
      "Iter 224: Loss=1.4289, Noise=0.9853\n",
      "Iter 225: Loss=1.4287, Noise=0.9855\n",
      "Iter 226: Loss=1.4287, Noise=0.9852\n",
      "Iter 227: Loss=1.4288, Noise=0.9838\n",
      "Iter 228: Loss=1.4286, Noise=0.9824\n",
      "Iter 229: Loss=1.4285, Noise=0.9809\n",
      "Iter 230: Loss=1.4288, Noise=0.9796\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 231: Loss=1.4289, Noise=0.9791\n",
      "Iter 232: Loss=1.4285, Noise=0.9789\n",
      "Iter 233: Loss=1.4289, Noise=0.9787\n",
      "Iter 234: Loss=1.4288, Noise=0.9787\n",
      "Iter 235: Loss=1.4288, Noise=0.9788\n",
      "Iter 236: Loss=1.4287, Noise=0.9798\n",
      "Iter 237: Loss=1.4287, Noise=0.9811\n",
      "Iter 238: Loss=1.4286, Noise=0.9822\n",
      "Iter 239: Loss=1.4288, Noise=0.9821\n",
      "Iter 240: Loss=1.4287, Noise=0.9817\n",
      "Iter 241: Loss=1.4288, Noise=0.9812\n",
      "Iter 242: Loss=1.4288, Noise=0.9818\n",
      "Iter 243: Loss=1.4286, Noise=0.9835\n",
      "Iter 244: Loss=1.4288, Noise=0.9853\n",
      "Iter 245: Loss=1.4287, Noise=0.9864\n",
      "Iter 246: Loss=1.4287, Noise=0.9877\n",
      "Iter 247: Loss=1.4287, Noise=0.9890\n",
      "Iter 248: Loss=1.4287, Noise=0.9896\n",
      "Iter 249: Loss=1.4289, Noise=0.9901\n",
      "Iter 250: Loss=1.4290, Noise=0.9903\n",
      "Iter 251: Loss=1.4288, Noise=0.9900\n",
      "Iter 252: Loss=1.4287, Noise=0.9896\n",
      "Iter 253: Loss=1.4288, Noise=0.9897\n",
      "Iter 254: Loss=1.4288, Noise=0.9896\n",
      "Iter 255: Loss=1.4285, Noise=0.9894\n",
      "Iter 256: Loss=1.4287, Noise=0.9896\n",
      "Iter 257: Loss=1.4288, Noise=0.9899\n",
      "Iter 258: Loss=1.4287, Noise=0.9902\n",
      "Iter 259: Loss=1.4287, Noise=0.9897\n",
      "Iter 260: Loss=1.4287, Noise=0.9885\n",
      "Iter 261: Loss=1.4287, Noise=0.9865\n",
      "Iter 262: Loss=1.4286, Noise=0.9853\n",
      "Iter 263: Loss=1.4287, Noise=0.9848\n",
      "Iter 264: Loss=1.4287, Noise=0.9837\n",
      "Iter 265: Loss=1.4286, Noise=0.9822\n",
      "Iter 266: Loss=1.4287, Noise=0.9819\n",
      "Iter 267: Loss=1.4288, Noise=0.9807\n",
      "Iter 268: Loss=1.4287, Noise=0.9802\n",
      "Iter 269: Loss=1.4285, Noise=0.9804\n",
      "Iter 270: Loss=1.4288, Noise=0.9813\n",
      "Iter 271: Loss=1.4289, Noise=0.9815\n",
      "Iter 272: Loss=1.4287, Noise=0.9811\n",
      "Iter 273: Loss=1.4287, Noise=0.9817\n",
      "Iter 274: Loss=1.4286, Noise=0.9828\n",
      "Iter 275: Loss=1.4286, Noise=0.9846\n",
      "Iter 276: Loss=1.4288, Noise=0.9859\n",
      "Iter 277: Loss=1.4287, Noise=0.9875\n",
      "Iter 278: Loss=1.4287, Noise=0.9880\n",
      "Iter 279: Loss=1.4286, Noise=0.9891\n",
      "Iter 280: Loss=1.4286, Noise=0.9909\n",
      "Iter 281: Loss=1.4286, Noise=0.9926\n",
      "Iter 282: Loss=1.4286, Noise=0.9931\n",
      "Iter 283: Loss=1.4288, Noise=0.9923\n",
      "Iter 284: Loss=1.4287, Noise=0.9911\n",
      "Iter 285: Loss=1.4286, Noise=0.9900\n",
      "Iter 286: Loss=1.4287, Noise=0.9882\n",
      "Iter 287: Loss=1.4287, Noise=0.9869\n",
      "Iter 288: Loss=1.4285, Noise=0.9857\n",
      "Iter 289: Loss=1.4288, Noise=0.9839\n",
      "Iter 290: Loss=1.4287, Noise=0.9814\n",
      "Iter 291: Loss=1.4287, Noise=0.9783\n",
      "Iter 292: Loss=1.4286, Noise=0.9766\n",
      "Iter 293: Loss=1.4286, Noise=0.9754\n",
      "Iter 294: Loss=1.4285, Noise=0.9738\n",
      "Iter 295: Loss=1.4287, Noise=0.9741\n",
      "Iter 296: Loss=1.4289, Noise=0.9743\n",
      "Iter 297: Loss=1.4287, Noise=0.9739\n",
      "Iter 298: Loss=1.4288, Noise=0.9733\n",
      "Iter 299: Loss=1.4287, Noise=0.9739\n",
      "Iter 300: Loss=1.4288, Noise=0.9753\n",
      "Iter 301: Loss=1.4288, Noise=0.9768\n",
      "Iter 302: Loss=1.4286, Noise=0.9784\n",
      "Iter 303: Loss=1.4286, Noise=0.9792\n",
      "Iter 304: Loss=1.4287, Noise=0.9806\n",
      "Iter 305: Loss=1.4286, Noise=0.9815\n",
      "Iter 306: Loss=1.4288, Noise=0.9838\n",
      "Iter 307: Loss=1.4285, Noise=0.9859\n",
      "Iter 308: Loss=1.4287, Noise=0.9872\n",
      "Iter 309: Loss=1.4289, Noise=0.9896\n",
      "Iter 310: Loss=1.4286, Noise=0.9910\n",
      "Iter 311: Loss=1.4286, Noise=0.9923\n",
      "Iter 312: Loss=1.4287, Noise=0.9924\n",
      "Iter 313: Loss=1.4288, Noise=0.9930\n",
      "Iter 314: Loss=1.4287, Noise=0.9925\n",
      "Iter 315: Loss=1.4288, Noise=0.9919\n",
      "Iter 316: Loss=1.4286, Noise=0.9914\n",
      "Iter 317: Loss=1.4287, Noise=0.9915\n",
      "Iter 318: Loss=1.4288, Noise=0.9916\n",
      "Iter 319: Loss=1.4288, Noise=0.9909\n",
      "Iter 320: Loss=1.4287, Noise=0.9905\n",
      "Iter 321: Loss=1.4288, Noise=0.9903\n",
      "Iter 322: Loss=1.4288, Noise=0.9901\n",
      "Iter 323: Loss=1.4288, Noise=0.9894\n",
      "Iter 324: Loss=1.4287, Noise=0.9894\n",
      "Iter 325: Loss=1.4288, Noise=0.9893\n",
      "Iter 326: Loss=1.4286, Noise=0.9896\n",
      "Iter 327: Loss=1.4287, Noise=0.9898\n",
      "Iter 328: Loss=1.4287, Noise=0.9896\n",
      "Iter 329: Loss=1.4286, Noise=0.9889\n",
      "Iter 330: Loss=1.4285, Noise=0.9883\n",
      "Iter 331: Loss=1.4287, Noise=0.9879\n",
      "Iter 332: Loss=1.4287, Noise=0.9868\n",
      "Iter 333: Loss=1.4288, Noise=0.9849\n",
      "Iter 334: Loss=1.4288, Noise=0.9821\n",
      "Iter 335: Loss=1.4285, Noise=0.9790\n",
      "Iter 336: Loss=1.4286, Noise=0.9758\n",
      "Iter 337: Loss=1.4287, Noise=0.9736\n",
      "Iter 338: Loss=1.4287, Noise=0.9722\n",
      "Iter 339: Loss=1.4288, Noise=0.9718\n",
      "Iter 340: Loss=1.4288, Noise=0.9712\n",
      "Iter 341: Loss=1.4289, Noise=0.9704\n",
      "Iter 342: Loss=1.4289, Noise=0.9695\n",
      "Iter 343: Loss=1.4288, Noise=0.9693\n",
      "Iter 344: Loss=1.4288, Noise=0.9699\n",
      "Iter 345: Loss=1.4288, Noise=0.9713\n",
      "Iter 346: Loss=1.4287, Noise=0.9722\n",
      "Iter 347: Loss=1.4286, Noise=0.9739\n",
      "Iter 348: Loss=1.4287, Noise=0.9753\n",
      "Iter 349: Loss=1.4287, Noise=0.9768\n",
      "Iter 350: Loss=1.4288, Noise=0.9789\n",
      "Iter 351: Loss=1.4287, Noise=0.9810\n",
      "Iter 352: Loss=1.4287, Noise=0.9834\n",
      "Iter 353: Loss=1.4289, Noise=0.9861\n",
      "Iter 354: Loss=1.4285, Noise=0.9880\n",
      "Iter 355: Loss=1.4288, Noise=0.9890\n",
      "Iter 356: Loss=1.4286, Noise=0.9887\n",
      "Iter 357: Loss=1.4287, Noise=0.9885\n",
      "Iter 358: Loss=1.4288, Noise=0.9885\n",
      "Iter 359: Loss=1.4288, Noise=0.9894\n",
      "Iter 360: Loss=1.4287, Noise=0.9909\n",
      "Iter 361: Loss=1.4287, Noise=0.9909\n",
      "Iter 362: Loss=1.4287, Noise=0.9904\n",
      "Iter 363: Loss=1.4286, Noise=0.9905\n",
      "Iter 364: Loss=1.4285, Noise=0.9902\n",
      "Iter 365: Loss=1.4287, Noise=0.9895\n",
      "Iter 366: Loss=1.4288, Noise=0.9894\n",
      "Iter 367: Loss=1.4288, Noise=0.9887\n",
      "Iter 368: Loss=1.4289, Noise=0.9881\n",
      "Iter 369: Loss=1.4288, Noise=0.9873\n",
      "Iter 370: Loss=1.4287, Noise=0.9865\n",
      "Iter 371: Loss=1.4287, Noise=0.9856\n",
      "Iter 372: Loss=1.4289, Noise=0.9843\n",
      "Iter 373: Loss=1.4286, Noise=0.9839\n",
      "Iter 374: Loss=1.4287, Noise=0.9834\n",
      "Iter 375: Loss=1.4287, Noise=0.9822\n",
      "Iter 376: Loss=1.4285, Noise=0.9811\n",
      "Iter 377: Loss=1.4288, Noise=0.9813\n",
      "Iter 378: Loss=1.4285, Noise=0.9811\n",
      "Iter 379: Loss=1.4287, Noise=0.9818\n",
      "Iter 380: Loss=1.4287, Noise=0.9828\n",
      "Iter 381: Loss=1.4288, Noise=0.9850\n",
      "Iter 382: Loss=1.4285, Noise=0.9865\n",
      "Iter 383: Loss=1.4287, Noise=0.9882\n",
      "Iter 384: Loss=1.4285, Noise=0.9900\n",
      "Iter 385: Loss=1.4287, Noise=0.9918\n",
      "Iter 386: Loss=1.4288, Noise=0.9932\n",
      "Iter 387: Loss=1.4283, Noise=0.9948\n",
      "Iter 388: Loss=1.4285, Noise=0.9955\n",
      "Iter 389: Loss=1.4288, Noise=0.9952\n",
      "Iter 390: Loss=1.4286, Noise=0.9950\n",
      "Iter 391: Loss=1.4287, Noise=0.9956\n",
      "Iter 392: Loss=1.4287, Noise=0.9956\n",
      "Iter 393: Loss=1.4287, Noise=0.9962\n",
      "Iter 394: Loss=1.4287, Noise=0.9984\n",
      "Iter 395: Loss=1.4287, Noise=1.0001\n",
      "Iter 396: Loss=1.4288, Noise=1.0013\n",
      "Iter 397: Loss=1.4286, Noise=1.0017\n",
      "Iter 398: Loss=1.4287, Noise=1.0013\n",
      "Iter 399: Loss=1.4287, Noise=1.0005\n",
      "Iter 400: Loss=1.4288, Noise=0.9988\n",
      "Iter 401: Loss=1.4286, Noise=0.9964\n",
      "Iter 402: Loss=1.4287, Noise=0.9927\n",
      "Iter 403: Loss=1.4286, Noise=0.9892\n",
      "Iter 404: Loss=1.4287, Noise=0.9860\n",
      "Iter 405: Loss=1.4286, Noise=0.9828\n",
      "Iter 406: Loss=1.4287, Noise=0.9801\n",
      "Iter 407: Loss=1.4285, Noise=0.9774\n",
      "Iter 408: Loss=1.4287, Noise=0.9747\n",
      "Iter 409: Loss=1.4288, Noise=0.9726\n",
      "Iter 410: Loss=1.4287, Noise=0.9718\n",
      "Iter 411: Loss=1.4287, Noise=0.9713\n",
      "Iter 412: Loss=1.4289, Noise=0.9715\n",
      "Iter 413: Loss=1.4290, Noise=0.9724\n",
      "Iter 414: Loss=1.4287, Noise=0.9737\n",
      "Iter 415: Loss=1.4288, Noise=0.9755\n",
      "Iter 416: Loss=1.4289, Noise=0.9772\n",
      "Iter 417: Loss=1.4288, Noise=0.9797\n",
      "Iter 418: Loss=1.4287, Noise=0.9819\n",
      "Iter 419: Loss=1.4287, Noise=0.9847\n",
      "Iter 420: Loss=1.4288, Noise=0.9867\n",
      "Iter 421: Loss=1.4287, Noise=0.9886\n",
      "Iter 422: Loss=1.4289, Noise=0.9904\n",
      "Iter 423: Loss=1.4286, Noise=0.9921\n",
      "Iter 424: Loss=1.4286, Noise=0.9938\n",
      "Iter 425: Loss=1.4287, Noise=0.9938\n",
      "Iter 426: Loss=1.4285, Noise=0.9924\n",
      "Iter 427: Loss=1.4288, Noise=0.9899\n",
      "Iter 428: Loss=1.4288, Noise=0.9875\n",
      "Iter 429: Loss=1.4287, Noise=0.9857\n",
      "Iter 430: Loss=1.4289, Noise=0.9835\n",
      "Iter 431: Loss=1.4287, Noise=0.9816\n",
      "Iter 432: Loss=1.4288, Noise=0.9794\n",
      "Iter 433: Loss=1.4285, Noise=0.9770\n",
      "Iter 434: Loss=1.4288, Noise=0.9736\n",
      "Iter 435: Loss=1.4288, Noise=0.9709\n",
      "Iter 436: Loss=1.4287, Noise=0.9691\n",
      "Iter 437: Loss=1.4287, Noise=0.9679\n",
      "Iter 438: Loss=1.4289, Noise=0.9669\n",
      "Iter 439: Loss=1.4288, Noise=0.9672\n",
      "Iter 440: Loss=1.4288, Noise=0.9676\n",
      "Iter 441: Loss=1.4288, Noise=0.9678\n",
      "Iter 442: Loss=1.4288, Noise=0.9692\n",
      "Iter 443: Loss=1.4289, Noise=0.9705\n",
      "Iter 444: Loss=1.4287, Noise=0.9711\n",
      "Iter 445: Loss=1.4287, Noise=0.9712\n",
      "Iter 446: Loss=1.4287, Noise=0.9726\n",
      "Iter 447: Loss=1.4288, Noise=0.9748\n",
      "Iter 448: Loss=1.4289, Noise=0.9775\n",
      "Iter 449: Loss=1.4287, Noise=0.9800\n",
      "Iter 450: Loss=1.4287, Noise=0.9823\n",
      "Iter 451: Loss=1.4285, Noise=0.9851\n",
      "Iter 452: Loss=1.4287, Noise=0.9875\n",
      "Iter 453: Loss=1.4286, Noise=0.9912\n",
      "Iter 454: Loss=1.4286, Noise=0.9955\n",
      "Iter 455: Loss=1.4288, Noise=0.9988\n",
      "Iter 456: Loss=1.4286, Noise=1.0016\n",
      "Iter 457: Loss=1.4288, Noise=1.0036\n",
      "Iter 458: Loss=1.4288, Noise=1.0043\n",
      "Iter 459: Loss=1.4289, Noise=1.0049\n",
      "Iter 460: Loss=1.4287, Noise=1.0049\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 461: Loss=1.4289, Noise=1.0047\n",
      "Iter 462: Loss=1.4287, Noise=1.0046\n",
      "Iter 463: Loss=1.4288, Noise=1.0050\n",
      "Iter 464: Loss=1.4286, Noise=1.0056\n",
      "Iter 465: Loss=1.4288, Noise=1.0059\n",
      "Iter 466: Loss=1.4287, Noise=1.0063\n",
      "Iter 467: Loss=1.4289, Noise=1.0067\n",
      "Iter 468: Loss=1.4289, Noise=1.0072\n",
      "Iter 469: Loss=1.4288, Noise=1.0068\n",
      "Iter 470: Loss=1.4287, Noise=1.0054\n",
      "Iter 471: Loss=1.4289, Noise=1.0046\n",
      "Iter 472: Loss=1.4288, Noise=1.0035\n",
      "Iter 473: Loss=1.4287, Noise=1.0016\n",
      "Iter 474: Loss=1.4288, Noise=1.0000\n",
      "Iter 475: Loss=1.4288, Noise=0.9980\n",
      "Iter 476: Loss=1.4288, Noise=0.9955\n",
      "Iter 477: Loss=1.4286, Noise=0.9938\n",
      "Iter 478: Loss=1.4287, Noise=0.9927\n",
      "Iter 479: Loss=1.4288, Noise=0.9908\n",
      "Iter 480: Loss=1.4287, Noise=0.9893\n",
      "Iter 481: Loss=1.4286, Noise=0.9878\n",
      "Iter 482: Loss=1.4288, Noise=0.9873\n",
      "Iter 483: Loss=1.4286, Noise=0.9875\n",
      "Iter 484: Loss=1.4286, Noise=0.9876\n",
      "Iter 485: Loss=1.4287, Noise=0.9887\n",
      "Iter 486: Loss=1.4287, Noise=0.9895\n",
      "Iter 487: Loss=1.4285, Noise=0.9898\n",
      "Iter 488: Loss=1.4288, Noise=0.9897\n",
      "Iter 489: Loss=1.4288, Noise=0.9909\n",
      "Iter 490: Loss=1.4288, Noise=0.9924\n",
      "Iter 491: Loss=1.4287, Noise=0.9928\n",
      "Iter 492: Loss=1.4287, Noise=0.9923\n",
      "Iter 493: Loss=1.4285, Noise=0.9907\n",
      "Iter 494: Loss=1.4286, Noise=0.9876\n",
      "Iter 495: Loss=1.4286, Noise=0.9845\n",
      "Iter 496: Loss=1.4288, Noise=0.9820\n",
      "Iter 497: Loss=1.4286, Noise=0.9796\n",
      "Iter 498: Loss=1.4287, Noise=0.9780\n",
      "Iter 499: Loss=1.4289, Noise=0.9775\n",
      "Iter 500: Loss=1.4289, Noise=0.9775\n",
      "Iter 501: Loss=1.4285, Noise=0.9780\n",
      "Iter 502: Loss=1.4286, Noise=0.9784\n",
      "Iter 503: Loss=1.4288, Noise=0.9797\n",
      "Iter 504: Loss=1.4287, Noise=0.9797\n",
      "Iter 505: Loss=1.4287, Noise=0.9803\n",
      "Iter 506: Loss=1.4286, Noise=0.9802\n",
      "Iter 507: Loss=1.4287, Noise=0.9798\n",
      "Iter 508: Loss=1.4287, Noise=0.9802\n",
      "Iter 509: Loss=1.4288, Noise=0.9801\n",
      "Iter 510: Loss=1.4287, Noise=0.9809\n",
      "Iter 511: Loss=1.4286, Noise=0.9808\n",
      "Iter 512: Loss=1.4288, Noise=0.9801\n",
      "Iter 513: Loss=1.4286, Noise=0.9786\n",
      "Iter 514: Loss=1.4288, Noise=0.9776\n",
      "Iter 515: Loss=1.4286, Noise=0.9769\n",
      "Iter 516: Loss=1.4287, Noise=0.9758\n",
      "Iter 517: Loss=1.4287, Noise=0.9749\n",
      "Iter 518: Loss=1.4288, Noise=0.9738\n",
      "Iter 519: Loss=1.4288, Noise=0.9725\n",
      "Iter 520: Loss=1.4286, Noise=0.9711\n",
      "Iter 521: Loss=1.4288, Noise=0.9707\n",
      "Iter 522: Loss=1.4287, Noise=0.9706\n",
      "Iter 523: Loss=1.4288, Noise=0.9709\n",
      "Iter 524: Loss=1.4288, Noise=0.9703\n",
      "Iter 525: Loss=1.4289, Noise=0.9702\n",
      "Iter 526: Loss=1.4288, Noise=0.9702\n",
      "Iter 527: Loss=1.4288, Noise=0.9711\n",
      "Iter 528: Loss=1.4289, Noise=0.9727\n",
      "Iter 529: Loss=1.4287, Noise=0.9746\n",
      "Iter 530: Loss=1.4287, Noise=0.9765\n",
      "Iter 531: Loss=1.4286, Noise=0.9793\n",
      "Iter 532: Loss=1.4287, Noise=0.9815\n",
      "Iter 533: Loss=1.4287, Noise=0.9841\n",
      "Iter 534: Loss=1.4288, Noise=0.9863\n",
      "Iter 535: Loss=1.4285, Noise=0.9878\n",
      "Iter 536: Loss=1.4288, Noise=0.9894\n",
      "Iter 537: Loss=1.4286, Noise=0.9910\n",
      "Iter 538: Loss=1.4287, Noise=0.9916\n",
      "Iter 539: Loss=1.4287, Noise=0.9925\n",
      "Iter 540: Loss=1.4287, Noise=0.9938\n",
      "Iter 541: Loss=1.4288, Noise=0.9938\n",
      "Iter 542: Loss=1.4287, Noise=0.9935\n",
      "Iter 543: Loss=1.4289, Noise=0.9931\n",
      "Iter 544: Loss=1.4288, Noise=0.9926\n",
      "Iter 545: Loss=1.4288, Noise=0.9925\n",
      "Iter 546: Loss=1.4287, Noise=0.9921\n",
      "Iter 547: Loss=1.4287, Noise=0.9917\n",
      "Iter 548: Loss=1.4287, Noise=0.9917\n",
      "Iter 549: Loss=1.4288, Noise=0.9917\n",
      "Iter 550: Loss=1.4288, Noise=0.9923\n",
      "Iter 551: Loss=1.4288, Noise=0.9930\n",
      "Iter 552: Loss=1.4287, Noise=0.9924\n",
      "Iter 553: Loss=1.4288, Noise=0.9921\n",
      "Iter 554: Loss=1.4288, Noise=0.9917\n",
      "Iter 555: Loss=1.4287, Noise=0.9909\n",
      "Iter 556: Loss=1.4287, Noise=0.9901\n",
      "Iter 557: Loss=1.4287, Noise=0.9890\n",
      "Iter 558: Loss=1.4287, Noise=0.9867\n",
      "Iter 559: Loss=1.4286, Noise=0.9848\n",
      "Iter 560: Loss=1.4285, Noise=0.9831\n",
      "Iter 561: Loss=1.4286, Noise=0.9803\n",
      "Iter 562: Loss=1.4286, Noise=0.9778\n",
      "Iter 563: Loss=1.4287, Noise=0.9764\n",
      "Iter 564: Loss=1.4286, Noise=0.9739\n",
      "Iter 565: Loss=1.4286, Noise=0.9713\n",
      "Iter 566: Loss=1.4288, Noise=0.9698\n",
      "Iter 567: Loss=1.4287, Noise=0.9693\n",
      "Iter 568: Loss=1.4289, Noise=0.9701\n",
      "Iter 569: Loss=1.4287, Noise=0.9710\n",
      "Iter 570: Loss=1.4288, Noise=0.9715\n",
      "Iter 571: Loss=1.4287, Noise=0.9737\n",
      "Iter 572: Loss=1.4286, Noise=0.9756\n",
      "Iter 573: Loss=1.4287, Noise=0.9771\n",
      "Iter 574: Loss=1.4286, Noise=0.9787\n",
      "Iter 575: Loss=1.4286, Noise=0.9796\n",
      "Iter 576: Loss=1.4285, Noise=0.9809\n",
      "Iter 577: Loss=1.4288, Noise=0.9827\n",
      "Iter 578: Loss=1.4289, Noise=0.9845\n",
      "Iter 579: Loss=1.4285, Noise=0.9864\n",
      "Iter 580: Loss=1.4287, Noise=0.9887\n",
      "Iter 581: Loss=1.4287, Noise=0.9907\n",
      "Iter 582: Loss=1.4287, Noise=0.9936\n",
      "Iter 583: Loss=1.4285, Noise=0.9963\n",
      "Iter 584: Loss=1.4288, Noise=0.9987\n",
      "Iter 585: Loss=1.4286, Noise=1.0013\n",
      "Iter 586: Loss=1.4287, Noise=1.0032\n",
      "Iter 587: Loss=1.4288, Noise=1.0055\n",
      "Iter 588: Loss=1.4289, Noise=1.0070\n",
      "Iter 589: Loss=1.4288, Noise=1.0077\n",
      "Iter 590: Loss=1.4288, Noise=1.0077\n",
      "Iter 591: Loss=1.4287, Noise=1.0072\n",
      "Iter 592: Loss=1.4286, Noise=1.0057\n",
      "Iter 593: Loss=1.4289, Noise=1.0046\n",
      "Iter 594: Loss=1.4288, Noise=1.0029\n",
      "Iter 595: Loss=1.4289, Noise=1.0006\n",
      "Iter 596: Loss=1.4288, Noise=0.9986\n",
      "Iter 597: Loss=1.4286, Noise=0.9976\n",
      "Iter 598: Loss=1.4287, Noise=0.9958\n",
      "Iter 599: Loss=1.4286, Noise=0.9936\n",
      "Iter 600: Loss=1.4288, Noise=0.9921\n",
      "Iter 601: Loss=1.4285, Noise=0.9908\n",
      "Iter 602: Loss=1.4287, Noise=0.9891\n",
      "Iter 603: Loss=1.4286, Noise=0.9865\n",
      "Iter 604: Loss=1.4289, Noise=0.9839\n",
      "Iter 605: Loss=1.4287, Noise=0.9807\n",
      "Iter 606: Loss=1.4287, Noise=0.9777\n",
      "Iter 607: Loss=1.4286, Noise=0.9764\n",
      "Iter 608: Loss=1.4287, Noise=0.9753\n",
      "Iter 609: Loss=1.4288, Noise=0.9750\n",
      "Iter 610: Loss=1.4288, Noise=0.9751\n",
      "Iter 611: Loss=1.4290, Noise=0.9751\n",
      "Iter 612: Loss=1.4287, Noise=0.9748\n",
      "Iter 613: Loss=1.4288, Noise=0.9740\n",
      "Iter 614: Loss=1.4287, Noise=0.9729\n",
      "Iter 615: Loss=1.4289, Noise=0.9715\n",
      "Iter 616: Loss=1.4286, Noise=0.9701\n",
      "Iter 617: Loss=1.4288, Noise=0.9693\n",
      "Iter 618: Loss=1.4288, Noise=0.9693\n",
      "Iter 619: Loss=1.4287, Noise=0.9701\n",
      "Iter 620: Loss=1.4287, Noise=0.9708\n",
      "Iter 621: Loss=1.4286, Noise=0.9718\n",
      "Iter 622: Loss=1.4287, Noise=0.9736\n",
      "Iter 623: Loss=1.4286, Noise=0.9760\n",
      "Iter 624: Loss=1.4287, Noise=0.9770\n",
      "Iter 625: Loss=1.4289, Noise=0.9785\n",
      "Iter 626: Loss=1.4289, Noise=0.9802\n",
      "Iter 627: Loss=1.4288, Noise=0.9828\n",
      "Iter 628: Loss=1.4287, Noise=0.9858\n",
      "Iter 629: Loss=1.4289, Noise=0.9882\n",
      "Iter 630: Loss=1.4288, Noise=0.9901\n",
      "Iter 631: Loss=1.4288, Noise=0.9915\n",
      "Iter 632: Loss=1.4287, Noise=0.9920\n",
      "Iter 633: Loss=1.4287, Noise=0.9909\n",
      "Iter 634: Loss=1.4287, Noise=0.9901\n",
      "Iter 635: Loss=1.4287, Noise=0.9898\n",
      "Iter 636: Loss=1.4287, Noise=0.9899\n",
      "Iter 637: Loss=1.4287, Noise=0.9902\n",
      "Iter 638: Loss=1.4288, Noise=0.9907\n",
      "Iter 639: Loss=1.4287, Noise=0.9915\n",
      "Iter 640: Loss=1.4287, Noise=0.9919\n",
      "Iter 641: Loss=1.4288, Noise=0.9910\n",
      "Iter 642: Loss=1.4287, Noise=0.9897\n",
      "Iter 643: Loss=1.4288, Noise=0.9897\n",
      "Iter 644: Loss=1.4288, Noise=0.9884\n",
      "Iter 645: Loss=1.4288, Noise=0.9867\n",
      "Iter 646: Loss=1.4287, Noise=0.9848\n",
      "Iter 647: Loss=1.4287, Noise=0.9837\n",
      "Iter 648: Loss=1.4287, Noise=0.9825\n",
      "Iter 649: Loss=1.4288, Noise=0.9806\n",
      "Iter 650: Loss=1.4288, Noise=0.9788\n",
      "Iter 651: Loss=1.4287, Noise=0.9781\n",
      "Iter 652: Loss=1.4286, Noise=0.9780\n",
      "Iter 653: Loss=1.4289, Noise=0.9780\n",
      "Iter 654: Loss=1.4287, Noise=0.9789\n",
      "Iter 655: Loss=1.4287, Noise=0.9805\n",
      "Iter 656: Loss=1.4287, Noise=0.9825\n",
      "Iter 657: Loss=1.4287, Noise=0.9854\n",
      "Iter 658: Loss=1.4288, Noise=0.9886\n",
      "Iter 659: Loss=1.4288, Noise=0.9912\n",
      "Iter 660: Loss=1.4286, Noise=0.9941\n",
      "Iter 661: Loss=1.4289, Noise=0.9964\n",
      "Iter 662: Loss=1.4288, Noise=0.9976\n",
      "Iter 663: Loss=1.4287, Noise=0.9982\n",
      "Iter 664: Loss=1.4289, Noise=0.9981\n",
      "Iter 665: Loss=1.4287, Noise=0.9990\n",
      "Iter 666: Loss=1.4286, Noise=0.9994\n",
      "Iter 667: Loss=1.4287, Noise=0.9989\n",
      "Iter 668: Loss=1.4286, Noise=0.9982\n",
      "Iter 669: Loss=1.4287, Noise=0.9976\n",
      "Iter 670: Loss=1.4289, Noise=0.9962\n",
      "Iter 671: Loss=1.4286, Noise=0.9945\n",
      "Iter 672: Loss=1.4288, Noise=0.9933\n",
      "Iter 673: Loss=1.4288, Noise=0.9909\n",
      "Iter 674: Loss=1.4289, Noise=0.9886\n",
      "Iter 675: Loss=1.4286, Noise=0.9863\n",
      "Iter 676: Loss=1.4286, Noise=0.9847\n",
      "Iter 677: Loss=1.4287, Noise=0.9822\n",
      "Iter 678: Loss=1.4288, Noise=0.9802\n",
      "Iter 679: Loss=1.4286, Noise=0.9780\n",
      "Iter 680: Loss=1.4287, Noise=0.9767\n",
      "Iter 681: Loss=1.4289, Noise=0.9753\n",
      "Iter 682: Loss=1.4288, Noise=0.9758\n",
      "Iter 683: Loss=1.4289, Noise=0.9759\n",
      "Iter 684: Loss=1.4286, Noise=0.9759\n",
      "Iter 685: Loss=1.4287, Noise=0.9765\n",
      "Iter 686: Loss=1.4286, Noise=0.9773\n",
      "Iter 687: Loss=1.4288, Noise=0.9783\n",
      "Iter 688: Loss=1.4287, Noise=0.9788\n",
      "Iter 689: Loss=1.4287, Noise=0.9799\n",
      "Iter 690: Loss=1.4288, Noise=0.9810\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 691: Loss=1.4288, Noise=0.9831\n",
      "Iter 692: Loss=1.4287, Noise=0.9856\n",
      "Iter 693: Loss=1.4287, Noise=0.9883\n",
      "Iter 694: Loss=1.4287, Noise=0.9905\n",
      "Iter 695: Loss=1.4285, Noise=0.9929\n",
      "Iter 696: Loss=1.4287, Noise=0.9936\n",
      "Iter 697: Loss=1.4288, Noise=0.9943\n",
      "Iter 698: Loss=1.4286, Noise=0.9951\n",
      "Iter 699: Loss=1.4288, Noise=0.9952\n",
      "Iter 700: Loss=1.4286, Noise=0.9949\n",
      "Iter 701: Loss=1.4288, Noise=0.9937\n",
      "Iter 702: Loss=1.4287, Noise=0.9920\n",
      "Iter 703: Loss=1.4287, Noise=0.9896\n",
      "Iter 704: Loss=1.4287, Noise=0.9865\n",
      "Iter 705: Loss=1.4286, Noise=0.9834\n",
      "Iter 706: Loss=1.4287, Noise=0.9808\n",
      "Iter 707: Loss=1.4287, Noise=0.9783\n",
      "Iter 708: Loss=1.4287, Noise=0.9756\n",
      "Iter 709: Loss=1.4287, Noise=0.9734\n",
      "Iter 710: Loss=1.4286, Noise=0.9726\n",
      "Iter 711: Loss=1.4287, Noise=0.9724\n",
      "Iter 712: Loss=1.4287, Noise=0.9721\n",
      "Iter 713: Loss=1.4286, Noise=0.9723\n",
      "Iter 714: Loss=1.4285, Noise=0.9733\n",
      "Iter 715: Loss=1.4287, Noise=0.9748\n",
      "Iter 716: Loss=1.4285, Noise=0.9770\n",
      "Iter 717: Loss=1.4287, Noise=0.9789\n",
      "Iter 718: Loss=1.4287, Noise=0.9816\n",
      "Iter 719: Loss=1.4288, Noise=0.9841\n",
      "Iter 720: Loss=1.4289, Noise=0.9862\n",
      "Iter 721: Loss=1.4288, Noise=0.9886\n",
      "Iter 722: Loss=1.4286, Noise=0.9902\n",
      "Iter 723: Loss=1.4289, Noise=0.9914\n",
      "Iter 724: Loss=1.4287, Noise=0.9926\n",
      "Iter 725: Loss=1.4286, Noise=0.9925\n",
      "Iter 726: Loss=1.4288, Noise=0.9928\n",
      "Iter 727: Loss=1.4287, Noise=0.9925\n",
      "Iter 728: Loss=1.4287, Noise=0.9933\n",
      "Iter 729: Loss=1.4288, Noise=0.9943\n",
      "Iter 730: Loss=1.4287, Noise=0.9952\n",
      "Iter 731: Loss=1.4288, Noise=0.9966\n",
      "Iter 732: Loss=1.4290, Noise=0.9977\n",
      "Iter 733: Loss=1.4288, Noise=0.9985\n",
      "Iter 734: Loss=1.4288, Noise=0.9991\n",
      "Iter 735: Loss=1.4288, Noise=1.0002\n",
      "Iter 736: Loss=1.4286, Noise=1.0010\n",
      "Iter 737: Loss=1.4287, Noise=1.0015\n",
      "Iter 738: Loss=1.4288, Noise=1.0002\n",
      "Iter 739: Loss=1.4287, Noise=0.9989\n",
      "Iter 740: Loss=1.4287, Noise=0.9974\n",
      "Iter 741: Loss=1.4286, Noise=0.9954\n",
      "Iter 742: Loss=1.4288, Noise=0.9931\n",
      "Iter 743: Loss=1.4289, Noise=0.9902\n",
      "Iter 744: Loss=1.4288, Noise=0.9879\n",
      "Iter 745: Loss=1.4286, Noise=0.9848\n",
      "Iter 746: Loss=1.4286, Noise=0.9821\n",
      "Iter 747: Loss=1.4288, Noise=0.9798\n",
      "Iter 748: Loss=1.4286, Noise=0.9781\n",
      "Iter 749: Loss=1.4286, Noise=0.9771\n",
      "Iter 750: Loss=1.4286, Noise=0.9753\n",
      "Iter 751: Loss=1.4285, Noise=0.9736\n",
      "Iter 752: Loss=1.4288, Noise=0.9735\n",
      "Iter 753: Loss=1.4287, Noise=0.9736\n",
      "Iter 754: Loss=1.4288, Noise=0.9738\n",
      "Iter 755: Loss=1.4287, Noise=0.9737\n",
      "Iter 756: Loss=1.4286, Noise=0.9737\n",
      "Iter 757: Loss=1.4285, Noise=0.9735\n",
      "Iter 758: Loss=1.4285, Noise=0.9733\n",
      "Iter 759: Loss=1.4289, Noise=0.9736\n",
      "Iter 760: Loss=1.4287, Noise=0.9737\n",
      "Iter 761: Loss=1.4287, Noise=0.9734\n",
      "Iter 762: Loss=1.4288, Noise=0.9732\n",
      "Iter 763: Loss=1.4286, Noise=0.9721\n",
      "Iter 764: Loss=1.4288, Noise=0.9711\n",
      "Iter 765: Loss=1.4287, Noise=0.9701\n",
      "Iter 766: Loss=1.4287, Noise=0.9696\n",
      "Iter 767: Loss=1.4288, Noise=0.9695\n",
      "Iter 768: Loss=1.4287, Noise=0.9706\n",
      "Iter 769: Loss=1.4286, Noise=0.9727\n",
      "Iter 770: Loss=1.4287, Noise=0.9750\n",
      "Iter 771: Loss=1.4287, Noise=0.9781\n",
      "Iter 772: Loss=1.4286, Noise=0.9808\n",
      "Iter 773: Loss=1.4286, Noise=0.9831\n",
      "Iter 774: Loss=1.4290, Noise=0.9851\n",
      "Iter 775: Loss=1.4286, Noise=0.9866\n",
      "Iter 776: Loss=1.4289, Noise=0.9870\n",
      "Iter 777: Loss=1.4286, Noise=0.9888\n",
      "Iter 778: Loss=1.4288, Noise=0.9902\n",
      "Iter 779: Loss=1.4287, Noise=0.9927\n",
      "Iter 780: Loss=1.4287, Noise=0.9945\n",
      "Iter 781: Loss=1.4288, Noise=0.9966\n",
      "Iter 782: Loss=1.4288, Noise=0.9985\n",
      "Iter 783: Loss=1.4288, Noise=0.9990\n",
      "Iter 784: Loss=1.4287, Noise=0.9989\n",
      "Iter 785: Loss=1.4286, Noise=0.9990\n",
      "Iter 786: Loss=1.4286, Noise=0.9983\n",
      "Iter 787: Loss=1.4286, Noise=0.9974\n",
      "Iter 788: Loss=1.4287, Noise=0.9960\n",
      "Iter 789: Loss=1.4286, Noise=0.9951\n",
      "Iter 790: Loss=1.4287, Noise=0.9935\n",
      "Iter 791: Loss=1.4287, Noise=0.9919\n",
      "Iter 792: Loss=1.4286, Noise=0.9905\n",
      "Iter 793: Loss=1.4287, Noise=0.9882\n",
      "Iter 794: Loss=1.4287, Noise=0.9861\n",
      "Iter 795: Loss=1.4288, Noise=0.9846\n",
      "Iter 796: Loss=1.4287, Noise=0.9832\n",
      "Iter 797: Loss=1.4287, Noise=0.9826\n",
      "Iter 798: Loss=1.4287, Noise=0.9819\n",
      "Iter 799: Loss=1.4288, Noise=0.9800\n",
      "Iter 800: Loss=1.4288, Noise=0.9782\n",
      "Iter 801: Loss=1.4286, Noise=0.9774\n",
      "Iter 802: Loss=1.4289, Noise=0.9771\n",
      "Iter 803: Loss=1.4288, Noise=0.9768\n",
      "Iter 804: Loss=1.4288, Noise=0.9758\n",
      "Iter 805: Loss=1.4287, Noise=0.9734\n",
      "Iter 806: Loss=1.4289, Noise=0.9714\n",
      "Iter 807: Loss=1.4288, Noise=0.9702\n",
      "Iter 808: Loss=1.4288, Noise=0.9694\n",
      "Iter 809: Loss=1.4287, Noise=0.9691\n",
      "Iter 810: Loss=1.4286, Noise=0.9702\n",
      "Iter 811: Loss=1.4287, Noise=0.9707\n",
      "Iter 812: Loss=1.4289, Noise=0.9718\n",
      "Iter 813: Loss=1.4288, Noise=0.9743\n",
      "Iter 814: Loss=1.4287, Noise=0.9767\n",
      "Iter 815: Loss=1.4285, Noise=0.9794\n",
      "Iter 816: Loss=1.4287, Noise=0.9820\n",
      "Iter 817: Loss=1.4287, Noise=0.9838\n",
      "Iter 818: Loss=1.4285, Noise=0.9856\n",
      "Iter 819: Loss=1.4287, Noise=0.9875\n",
      "Iter 820: Loss=1.4287, Noise=0.9881\n",
      "Iter 821: Loss=1.4287, Noise=0.9888\n",
      "Iter 822: Loss=1.4287, Noise=0.9888\n",
      "Iter 823: Loss=1.4286, Noise=0.9893\n",
      "Iter 824: Loss=1.4286, Noise=0.9891\n",
      "Iter 825: Loss=1.4287, Noise=0.9883\n",
      "Iter 826: Loss=1.4287, Noise=0.9879\n",
      "Iter 827: Loss=1.4288, Noise=0.9871\n",
      "Iter 828: Loss=1.4288, Noise=0.9871\n",
      "Iter 829: Loss=1.4289, Noise=0.9872\n",
      "Iter 830: Loss=1.4288, Noise=0.9871\n",
      "Iter 831: Loss=1.4288, Noise=0.9873\n",
      "Iter 832: Loss=1.4287, Noise=0.9883\n",
      "Iter 833: Loss=1.4288, Noise=0.9895\n",
      "Iter 834: Loss=1.4288, Noise=0.9910\n",
      "Iter 835: Loss=1.4288, Noise=0.9922\n",
      "Iter 836: Loss=1.4288, Noise=0.9926\n",
      "Iter 837: Loss=1.4287, Noise=0.9930\n",
      "Iter 838: Loss=1.4289, Noise=0.9936\n",
      "Iter 839: Loss=1.4286, Noise=0.9943\n",
      "Iter 840: Loss=1.4288, Noise=0.9953\n",
      "Iter 841: Loss=1.4286, Noise=0.9963\n",
      "Iter 842: Loss=1.4288, Noise=0.9975\n",
      "Iter 843: Loss=1.4288, Noise=0.9973\n",
      "Iter 844: Loss=1.4287, Noise=0.9973\n",
      "Iter 845: Loss=1.4287, Noise=0.9967\n",
      "Iter 846: Loss=1.4287, Noise=0.9961\n",
      "Iter 847: Loss=1.4287, Noise=0.9954\n",
      "Iter 848: Loss=1.4287, Noise=0.9948\n",
      "Iter 849: Loss=1.4287, Noise=0.9936\n",
      "Iter 850: Loss=1.4288, Noise=0.9924\n",
      "Iter 851: Loss=1.4287, Noise=0.9913\n",
      "Iter 852: Loss=1.4289, Noise=0.9904\n",
      "Iter 853: Loss=1.4285, Noise=0.9889\n",
      "Iter 854: Loss=1.4287, Noise=0.9868\n",
      "Iter 855: Loss=1.4288, Noise=0.9855\n",
      "Iter 856: Loss=1.4285, Noise=0.9841\n",
      "Iter 857: Loss=1.4288, Noise=0.9834\n",
      "Iter 858: Loss=1.4286, Noise=0.9836\n",
      "Iter 859: Loss=1.4288, Noise=0.9843\n",
      "Iter 860: Loss=1.4288, Noise=0.9846\n",
      "Iter 861: Loss=1.4286, Noise=0.9849\n",
      "Iter 862: Loss=1.4288, Noise=0.9858\n",
      "Iter 863: Loss=1.4289, Noise=0.9873\n",
      "Iter 864: Loss=1.4286, Noise=0.9878\n",
      "Iter 865: Loss=1.4287, Noise=0.9883\n",
      "Iter 866: Loss=1.4287, Noise=0.9889\n",
      "Iter 867: Loss=1.4286, Noise=0.9904\n",
      "Iter 868: Loss=1.4286, Noise=0.9901\n",
      "Iter 869: Loss=1.4288, Noise=0.9897\n",
      "Iter 870: Loss=1.4288, Noise=0.9887\n",
      "Iter 871: Loss=1.4286, Noise=0.9885\n",
      "Iter 872: Loss=1.4287, Noise=0.9886\n",
      "Iter 873: Loss=1.4287, Noise=0.9891\n",
      "Iter 874: Loss=1.4289, Noise=0.9900\n",
      "Iter 875: Loss=1.4286, Noise=0.9907\n",
      "Iter 876: Loss=1.4287, Noise=0.9907\n",
      "Iter 877: Loss=1.4289, Noise=0.9903\n",
      "Iter 878: Loss=1.4289, Noise=0.9897\n",
      "Iter 879: Loss=1.4288, Noise=0.9889\n",
      "Iter 880: Loss=1.4285, Noise=0.9893\n",
      "Iter 881: Loss=1.4287, Noise=0.9887\n",
      "Iter 882: Loss=1.4288, Noise=0.9879\n",
      "Iter 883: Loss=1.4287, Noise=0.9858\n",
      "Iter 884: Loss=1.4288, Noise=0.9842\n",
      "Iter 885: Loss=1.4286, Noise=0.9832\n",
      "Iter 886: Loss=1.4286, Noise=0.9826\n",
      "Iter 887: Loss=1.4288, Noise=0.9830\n",
      "Iter 888: Loss=1.4288, Noise=0.9832\n",
      "Iter 889: Loss=1.4288, Noise=0.9832\n",
      "Iter 890: Loss=1.4287, Noise=0.9843\n",
      "Iter 891: Loss=1.4288, Noise=0.9860\n",
      "Iter 892: Loss=1.4287, Noise=0.9882\n",
      "Iter 893: Loss=1.4288, Noise=0.9897\n",
      "Iter 894: Loss=1.4287, Noise=0.9913\n",
      "Iter 895: Loss=1.4288, Noise=0.9928\n",
      "Iter 896: Loss=1.4288, Noise=0.9939\n",
      "Iter 897: Loss=1.4287, Noise=0.9942\n",
      "Iter 898: Loss=1.4286, Noise=0.9945\n",
      "Iter 899: Loss=1.4288, Noise=0.9945\n",
      "Iter 900: Loss=1.4287, Noise=0.9935\n",
      "Iter 901: Loss=1.4288, Noise=0.9924\n",
      "Iter 902: Loss=1.4286, Noise=0.9920\n",
      "Iter 903: Loss=1.4286, Noise=0.9913\n",
      "Iter 904: Loss=1.4286, Noise=0.9906\n",
      "Iter 905: Loss=1.4287, Noise=0.9901\n",
      "Iter 906: Loss=1.4286, Noise=0.9889\n",
      "Iter 907: Loss=1.4286, Noise=0.9871\n",
      "Iter 908: Loss=1.4286, Noise=0.9863\n",
      "Iter 909: Loss=1.4286, Noise=0.9861\n",
      "Iter 910: Loss=1.4288, Noise=0.9854\n",
      "Iter 911: Loss=1.4284, Noise=0.9849\n",
      "Iter 912: Loss=1.4286, Noise=0.9846\n",
      "Iter 913: Loss=1.4285, Noise=0.9839\n",
      "Iter 914: Loss=1.4288, Noise=0.9836\n",
      "Iter 915: Loss=1.4288, Noise=0.9838\n",
      "Iter 916: Loss=1.4288, Noise=0.9838\n",
      "Iter 917: Loss=1.4287, Noise=0.9841\n",
      "Iter 918: Loss=1.4288, Noise=0.9846\n",
      "Iter 919: Loss=1.4288, Noise=0.9849\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 920: Loss=1.4287, Noise=0.9853\n",
      "Iter 921: Loss=1.4287, Noise=0.9867\n",
      "Iter 922: Loss=1.4287, Noise=0.9875\n",
      "Iter 923: Loss=1.4288, Noise=0.9884\n",
      "Iter 924: Loss=1.4287, Noise=0.9894\n",
      "Iter 925: Loss=1.4286, Noise=0.9902\n",
      "Iter 926: Loss=1.4289, Noise=0.9901\n",
      "Iter 927: Loss=1.4286, Noise=0.9907\n",
      "Iter 928: Loss=1.4287, Noise=0.9909\n",
      "Iter 929: Loss=1.4286, Noise=0.9902\n",
      "Iter 930: Loss=1.4286, Noise=0.9883\n",
      "Iter 931: Loss=1.4288, Noise=0.9868\n",
      "Iter 932: Loss=1.4287, Noise=0.9859\n",
      "Iter 933: Loss=1.4287, Noise=0.9859\n",
      "Iter 934: Loss=1.4289, Noise=0.9850\n",
      "Iter 935: Loss=1.4284, Noise=0.9838\n",
      "Iter 936: Loss=1.4287, Noise=0.9824\n",
      "Iter 937: Loss=1.4288, Noise=0.9815\n",
      "Iter 938: Loss=1.4288, Noise=0.9813\n",
      "Iter 939: Loss=1.4287, Noise=0.9824\n",
      "Iter 940: Loss=1.4287, Noise=0.9841\n",
      "Iter 941: Loss=1.4287, Noise=0.9874\n",
      "Iter 942: Loss=1.4288, Noise=0.9903\n",
      "Iter 943: Loss=1.4288, Noise=0.9935\n",
      "Iter 944: Loss=1.4287, Noise=0.9969\n",
      "Iter 945: Loss=1.4286, Noise=0.9990\n",
      "Iter 946: Loss=1.4289, Noise=1.0020\n",
      "Iter 947: Loss=1.4286, Noise=1.0041\n",
      "Iter 948: Loss=1.4287, Noise=1.0056\n",
      "Iter 949: Loss=1.4287, Noise=1.0060\n",
      "Iter 950: Loss=1.4289, Noise=1.0063\n",
      "Iter 951: Loss=1.4289, Noise=1.0067\n",
      "Iter 952: Loss=1.4288, Noise=1.0051\n",
      "Iter 953: Loss=1.4288, Noise=1.0027\n",
      "Iter 954: Loss=1.4287, Noise=1.0002\n",
      "Iter 955: Loss=1.4289, Noise=0.9975\n",
      "Iter 956: Loss=1.4286, Noise=0.9949\n",
      "Iter 957: Loss=1.4288, Noise=0.9921\n",
      "Iter 958: Loss=1.4286, Noise=0.9902\n",
      "Iter 959: Loss=1.4286, Noise=0.9878\n",
      "Iter 960: Loss=1.4286, Noise=0.9858\n",
      "Iter 961: Loss=1.4288, Noise=0.9837\n",
      "Iter 962: Loss=1.4289, Noise=0.9814\n",
      "Iter 963: Loss=1.4287, Noise=0.9784\n",
      "Iter 964: Loss=1.4287, Noise=0.9752\n",
      "Iter 965: Loss=1.4288, Noise=0.9715\n",
      "Iter 966: Loss=1.4289, Noise=0.9684\n",
      "Iter 967: Loss=1.4287, Noise=0.9656\n",
      "Iter 968: Loss=1.4288, Noise=0.9649\n",
      "Iter 969: Loss=1.4288, Noise=0.9644\n",
      "Iter 970: Loss=1.4288, Noise=0.9648\n",
      "Iter 971: Loss=1.4287, Noise=0.9663\n",
      "Iter 972: Loss=1.4289, Noise=0.9685\n",
      "Iter 973: Loss=1.4286, Noise=0.9713\n",
      "Iter 974: Loss=1.4288, Noise=0.9744\n",
      "Iter 975: Loss=1.4288, Noise=0.9779\n",
      "Iter 976: Loss=1.4287, Noise=0.9815\n",
      "Iter 977: Loss=1.4289, Noise=0.9840\n",
      "Iter 978: Loss=1.4286, Noise=0.9858\n",
      "Iter 979: Loss=1.4286, Noise=0.9869\n",
      "Iter 980: Loss=1.4287, Noise=0.9882\n",
      "Iter 981: Loss=1.4287, Noise=0.9896\n",
      "Iter 982: Loss=1.4287, Noise=0.9904\n",
      "Iter 983: Loss=1.4287, Noise=0.9902\n",
      "Iter 984: Loss=1.4287, Noise=0.9903\n",
      "Iter 985: Loss=1.4287, Noise=0.9899\n",
      "Iter 986: Loss=1.4288, Noise=0.9900\n",
      "Iter 987: Loss=1.4288, Noise=0.9902\n",
      "Iter 988: Loss=1.4288, Noise=0.9905\n",
      "Iter 989: Loss=1.4286, Noise=0.9904\n",
      "Iter 990: Loss=1.4288, Noise=0.9908\n",
      "Iter 991: Loss=1.4286, Noise=0.9908\n",
      "Iter 992: Loss=1.4288, Noise=0.9905\n",
      "Iter 993: Loss=1.4287, Noise=0.9894\n",
      "Iter 994: Loss=1.4287, Noise=0.9897\n",
      "Iter 995: Loss=1.4287, Noise=0.9899\n",
      "Iter 996: Loss=1.4286, Noise=0.9886\n",
      "Iter 997: Loss=1.4287, Noise=0.9876\n",
      "Iter 998: Loss=1.4286, Noise=0.9869\n",
      "Iter 999: Loss=1.4285, Noise=0.9877\n"
     ]
    }
   ],
   "source": [
    "# Train the model\n",
    "mll = ExactMarginalLogLikelihood(model.likelihood, model)\n",
    "mll.train()\n",
    "optimizer = Adam(mll.parameters(), lr=0.01)\n",
    "for iteration in range(1000):\n",
    "    optimizer.zero_grad()\n",
    "    output = model(train_x)\n",
    "    loss = -mll(output, train_y)\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "    msg = \"Iter {}: Loss={:2.4f}, Noise={:2.4f}\".format(iteration, loss.item(), model.likelihood.noise.item())\n",
    "    print(msg)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSE=0.9146, Neg Log Likelihood=1.3390\n"
     ]
    }
   ],
   "source": [
    "# Test the model\n",
    "model.eval()\n",
    "output = model(test_x)\n",
    "rmse = (test_y - output.mean).pow(2).mean().sqrt().item()\n",
    "neg_log_lik = -mll(output, test_y).item()\n",
    "print('RMSE={:2.4f}, Neg Log Likelihood={:2.4f}'.format(rmse, neg_log_lik))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create an DPA-GP-ARD model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Draw random projections, `diversify` and store in a linear module\n",
    "# Here, we are drawing 20 Gaussian projections into 1 dimension.\n",
    "projs = [rp.gen_rp(d, 1, dist='gaussian') for _ in range(num_projs)] # initial directions\n",
    "\n",
    "newW, _ = rp.space_equally(torch.cat(projs,dim=1).t(), lr=0.1, niter=5000) # Try to diversify\n",
    "newW.requires_grad = False # Make sure they aren't trainable\n",
    "projs = [newW[i:i+1, :].t() for i in range(0, num_projs, 1)]  # Reshape like initial directions\n",
    "\n",
    "proj_module = torch.nn.Linear(d, num_projs, bias=False)\n",
    "proj_module.weight.data = torch.cat(projs, dim=1).t()\n",
    "\n",
    "# Create the additive model that operates over these projections\n",
    "# Fixing the outputscale and lengthscale of the base kernels.\n",
    "base_kernel = RBFKernel()\n",
    "base_kernel.initialize(lengthscale=torch.tensor([1.]))\n",
    "base_kernel = ScaleKernel(base_kernel)\n",
    "base_kernel.initialize(outputscale=torch.tensor([1/num_projs]))\n",
    "\n",
    "# Combine into a single module.\n",
    "# Using prescale=True applies lengthscales to original input space, i.e. scaling pre-projection.\n",
    "kernel = ScaledProjectionKernel(proj_module, base_kernel, \n",
    "                                prescale=True,  \n",
    "                                ard_num_dims=d,\n",
    "                                learn_proj=False)\n",
    "# Or, just call the method from training_routines that wraps this initialization\n",
    "# from training_routines import create_additive_rp_kernel\n",
    "# create_additive_rp_kernel(d, num_projs, learn_proj=False, kernel_type='RBF', \n",
    "#                           space_proj=True, prescale=True, ard=True, k=1, \n",
    "#                           proj_dist='gaussian')\n",
    "\n",
    "kernel = ScaleKernel(kernel) # Optionally wrap with an additional ScaleKernel \n",
    "\n",
    "# Create an ExactGP model with this kernel\n",
    "likelihood = GaussianLikelihood()\n",
    "likelihood.noise = 1.\n",
    "model = ExactGPModel(train_x, train_y, GaussianLikelihood(), kernel)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 0: Loss=1.4611, Noise=0.6983\n",
      "Iter 1: Loss=1.4591, Noise=0.7033\n",
      "Iter 2: Loss=1.4581, Noise=0.7084\n",
      "Iter 3: Loss=1.4563, Noise=0.7134\n",
      "Iter 4: Loss=1.4553, Noise=0.7185\n",
      "Iter 5: Loss=1.4539, Noise=0.7236\n",
      "Iter 6: Loss=1.4528, Noise=0.7287\n",
      "Iter 7: Loss=1.4515, Noise=0.7337\n",
      "Iter 8: Loss=1.4502, Noise=0.7387\n",
      "Iter 9: Loss=1.4492, Noise=0.7437\n",
      "Iter 10: Loss=1.4483, Noise=0.7486\n",
      "Iter 11: Loss=1.4474, Noise=0.7535\n",
      "Iter 12: Loss=1.4466, Noise=0.7584\n",
      "Iter 13: Loss=1.4452, Noise=0.7633\n",
      "Iter 14: Loss=1.4446, Noise=0.7682\n",
      "Iter 15: Loss=1.4438, Noise=0.7731\n",
      "Iter 16: Loss=1.4430, Noise=0.7779\n",
      "Iter 17: Loss=1.4421, Noise=0.7827\n",
      "Iter 18: Loss=1.4414, Noise=0.7875\n",
      "Iter 19: Loss=1.4406, Noise=0.7921\n",
      "Iter 20: Loss=1.4395, Noise=0.7969\n",
      "Iter 21: Loss=1.4391, Noise=0.8015\n",
      "Iter 22: Loss=1.4388, Noise=0.8062\n",
      "Iter 23: Loss=1.4380, Noise=0.8109\n",
      "Iter 24: Loss=1.4374, Noise=0.8155\n",
      "Iter 25: Loss=1.4368, Noise=0.8201\n",
      "Iter 26: Loss=1.4362, Noise=0.8247\n",
      "Iter 27: Loss=1.4358, Noise=0.8292\n",
      "Iter 28: Loss=1.4351, Noise=0.8336\n",
      "Iter 29: Loss=1.4350, Noise=0.8379\n",
      "Iter 30: Loss=1.4344, Noise=0.8421\n",
      "Iter 31: Loss=1.4343, Noise=0.8462\n",
      "Iter 32: Loss=1.4334, Noise=0.8503\n",
      "Iter 33: Loss=1.4334, Noise=0.8543\n",
      "Iter 34: Loss=1.4331, Noise=0.8583\n",
      "Iter 35: Loss=1.4327, Noise=0.8621\n",
      "Iter 36: Loss=1.4324, Noise=0.8658\n",
      "Iter 37: Loss=1.4322, Noise=0.8695\n",
      "Iter 38: Loss=1.4320, Noise=0.8730\n",
      "Iter 39: Loss=1.4317, Noise=0.8766\n",
      "Iter 40: Loss=1.4314, Noise=0.8801\n",
      "Iter 41: Loss=1.4311, Noise=0.8834\n",
      "Iter 42: Loss=1.4307, Noise=0.8866\n",
      "Iter 43: Loss=1.4308, Noise=0.8897\n",
      "Iter 44: Loss=1.4306, Noise=0.8927\n",
      "Iter 45: Loss=1.4306, Noise=0.8956\n",
      "Iter 46: Loss=1.4302, Noise=0.8986\n",
      "Iter 47: Loss=1.4300, Noise=0.9015\n",
      "Iter 48: Loss=1.4300, Noise=0.9045\n",
      "Iter 49: Loss=1.4301, Noise=0.9074\n",
      "Iter 50: Loss=1.4298, Noise=0.9101\n",
      "Iter 51: Loss=1.4298, Noise=0.9127\n",
      "Iter 52: Loss=1.4295, Noise=0.9153\n",
      "Iter 53: Loss=1.4294, Noise=0.9177\n",
      "Iter 54: Loss=1.4294, Noise=0.9200\n",
      "Iter 55: Loss=1.4293, Noise=0.9222\n",
      "Iter 56: Loss=1.4294, Noise=0.9244\n",
      "Iter 57: Loss=1.4295, Noise=0.9266\n",
      "Iter 58: Loss=1.4294, Noise=0.9287\n",
      "Iter 59: Loss=1.4292, Noise=0.9308\n",
      "Iter 60: Loss=1.4291, Noise=0.9330\n",
      "Iter 61: Loss=1.4292, Noise=0.9349\n",
      "Iter 62: Loss=1.4290, Noise=0.9368\n",
      "Iter 63: Loss=1.4288, Noise=0.9386\n",
      "Iter 64: Loss=1.4289, Noise=0.9403\n",
      "Iter 65: Loss=1.4287, Noise=0.9421\n",
      "Iter 66: Loss=1.4288, Noise=0.9438\n",
      "Iter 67: Loss=1.4291, Noise=0.9455\n",
      "Iter 68: Loss=1.4285, Noise=0.9471\n",
      "Iter 69: Loss=1.4289, Noise=0.9487\n",
      "Iter 70: Loss=1.4289, Noise=0.9502\n",
      "Iter 71: Loss=1.4287, Noise=0.9517\n",
      "Iter 72: Loss=1.4290, Noise=0.9532\n",
      "Iter 73: Loss=1.4286, Noise=0.9546\n",
      "Iter 74: Loss=1.4286, Noise=0.9560\n",
      "Iter 75: Loss=1.4289, Noise=0.9573\n",
      "Iter 76: Loss=1.4289, Noise=0.9585\n",
      "Iter 77: Loss=1.4284, Noise=0.9598\n",
      "Iter 78: Loss=1.4288, Noise=0.9609\n",
      "Iter 79: Loss=1.4285, Noise=0.9619\n",
      "Iter 80: Loss=1.4285, Noise=0.9626\n",
      "Iter 81: Loss=1.4287, Noise=0.9632\n",
      "Iter 82: Loss=1.4287, Noise=0.9638\n",
      "Iter 83: Loss=1.4288, Noise=0.9642\n",
      "Iter 84: Loss=1.4283, Noise=0.9647\n",
      "Iter 85: Loss=1.4288, Noise=0.9652\n",
      "Iter 86: Loss=1.4288, Noise=0.9657\n",
      "Iter 87: Loss=1.4285, Noise=0.9662\n",
      "Iter 88: Loss=1.4285, Noise=0.9665\n",
      "Iter 89: Loss=1.4283, Noise=0.9670\n",
      "Iter 90: Loss=1.4284, Noise=0.9675\n",
      "Iter 91: Loss=1.4286, Noise=0.9681\n",
      "Iter 92: Loss=1.4284, Noise=0.9686\n",
      "Iter 93: Loss=1.4283, Noise=0.9691\n",
      "Iter 94: Loss=1.4286, Noise=0.9695\n",
      "Iter 95: Loss=1.4283, Noise=0.9700\n",
      "Iter 96: Loss=1.4287, Noise=0.9705\n",
      "Iter 97: Loss=1.4285, Noise=0.9710\n",
      "Iter 98: Loss=1.4282, Noise=0.9713\n",
      "Iter 99: Loss=1.4284, Noise=0.9715\n",
      "Iter 100: Loss=1.4284, Noise=0.9717\n",
      "Iter 101: Loss=1.4285, Noise=0.9720\n",
      "Iter 102: Loss=1.4287, Noise=0.9723\n",
      "Iter 103: Loss=1.4285, Noise=0.9725\n",
      "Iter 104: Loss=1.4286, Noise=0.9728\n",
      "Iter 105: Loss=1.4283, Noise=0.9732\n",
      "Iter 106: Loss=1.4280, Noise=0.9736\n",
      "Iter 107: Loss=1.4284, Noise=0.9740\n",
      "Iter 108: Loss=1.4283, Noise=0.9745\n",
      "Iter 109: Loss=1.4284, Noise=0.9749\n",
      "Iter 110: Loss=1.4282, Noise=0.9753\n",
      "Iter 111: Loss=1.4285, Noise=0.9756\n",
      "Iter 112: Loss=1.4281, Noise=0.9759\n",
      "Iter 113: Loss=1.4283, Noise=0.9761\n",
      "Iter 114: Loss=1.4280, Noise=0.9760\n",
      "Iter 115: Loss=1.4284, Noise=0.9758\n",
      "Iter 116: Loss=1.4282, Noise=0.9756\n",
      "Iter 117: Loss=1.4282, Noise=0.9756\n",
      "Iter 118: Loss=1.4280, Noise=0.9756\n",
      "Iter 119: Loss=1.4282, Noise=0.9756\n",
      "Iter 120: Loss=1.4280, Noise=0.9755\n",
      "Iter 121: Loss=1.4281, Noise=0.9753\n",
      "Iter 122: Loss=1.4279, Noise=0.9750\n",
      "Iter 123: Loss=1.4280, Noise=0.9747\n",
      "Iter 124: Loss=1.4279, Noise=0.9742\n",
      "Iter 125: Loss=1.4277, Noise=0.9738\n",
      "Iter 126: Loss=1.4279, Noise=0.9735\n",
      "Iter 127: Loss=1.4276, Noise=0.9731\n",
      "Iter 128: Loss=1.4274, Noise=0.9725\n",
      "Iter 129: Loss=1.4279, Noise=0.9719\n",
      "Iter 130: Loss=1.4277, Noise=0.9713\n",
      "Iter 131: Loss=1.4276, Noise=0.9707\n",
      "Iter 132: Loss=1.4276, Noise=0.9700\n",
      "Iter 133: Loss=1.4273, Noise=0.9693\n",
      "Iter 134: Loss=1.4272, Noise=0.9688\n",
      "Iter 135: Loss=1.4274, Noise=0.9685\n",
      "Iter 136: Loss=1.4273, Noise=0.9684\n",
      "Iter 137: Loss=1.4273, Noise=0.9684\n",
      "Iter 138: Loss=1.4273, Noise=0.9684\n",
      "Iter 139: Loss=1.4273, Noise=0.9683\n",
      "Iter 140: Loss=1.4271, Noise=0.9682\n",
      "Iter 141: Loss=1.4268, Noise=0.9682\n",
      "Iter 142: Loss=1.4268, Noise=0.9680\n",
      "Iter 143: Loss=1.4271, Noise=0.9679\n",
      "Iter 144: Loss=1.4271, Noise=0.9678\n",
      "Iter 145: Loss=1.4264, Noise=0.9676\n",
      "Iter 146: Loss=1.4266, Noise=0.9674\n",
      "Iter 147: Loss=1.4262, Noise=0.9674\n",
      "Iter 148: Loss=1.4267, Noise=0.9674\n",
      "Iter 149: Loss=1.4261, Noise=0.9675\n",
      "Iter 150: Loss=1.4264, Noise=0.9672\n",
      "Iter 151: Loss=1.4261, Noise=0.9668\n",
      "Iter 152: Loss=1.4261, Noise=0.9666\n",
      "Iter 153: Loss=1.4260, Noise=0.9663\n",
      "Iter 154: Loss=1.4256, Noise=0.9660\n",
      "Iter 155: Loss=1.4260, Noise=0.9655\n",
      "Iter 156: Loss=1.4257, Noise=0.9651\n",
      "Iter 157: Loss=1.4257, Noise=0.9648\n",
      "Iter 158: Loss=1.4255, Noise=0.9643\n",
      "Iter 159: Loss=1.4252, Noise=0.9641\n",
      "Iter 160: Loss=1.4252, Noise=0.9637\n",
      "Iter 161: Loss=1.4252, Noise=0.9632\n",
      "Iter 162: Loss=1.4246, Noise=0.9627\n",
      "Iter 163: Loss=1.4248, Noise=0.9621\n",
      "Iter 164: Loss=1.4249, Noise=0.9614\n",
      "Iter 165: Loss=1.4239, Noise=0.9607\n",
      "Iter 166: Loss=1.4242, Noise=0.9601\n",
      "Iter 167: Loss=1.4241, Noise=0.9595\n",
      "Iter 168: Loss=1.4243, Noise=0.9588\n",
      "Iter 169: Loss=1.4237, Noise=0.9581\n",
      "Iter 170: Loss=1.4234, Noise=0.9574\n",
      "Iter 171: Loss=1.4236, Noise=0.9567\n",
      "Iter 172: Loss=1.4230, Noise=0.9559\n",
      "Iter 173: Loss=1.4228, Noise=0.9551\n",
      "Iter 174: Loss=1.4229, Noise=0.9543\n",
      "Iter 175: Loss=1.4225, Noise=0.9535\n",
      "Iter 176: Loss=1.4223, Noise=0.9527\n",
      "Iter 177: Loss=1.4226, Noise=0.9521\n",
      "Iter 178: Loss=1.4222, Noise=0.9511\n",
      "Iter 179: Loss=1.4218, Noise=0.9502\n",
      "Iter 180: Loss=1.4214, Noise=0.9493\n",
      "Iter 181: Loss=1.4210, Noise=0.9480\n",
      "Iter 182: Loss=1.4211, Noise=0.9466\n",
      "Iter 183: Loss=1.4207, Noise=0.9454\n",
      "Iter 184: Loss=1.4209, Noise=0.9442\n",
      "Iter 185: Loss=1.4205, Noise=0.9430\n",
      "Iter 186: Loss=1.4199, Noise=0.9420\n",
      "Iter 187: Loss=1.4201, Noise=0.9408\n",
      "Iter 188: Loss=1.4199, Noise=0.9396\n",
      "Iter 189: Loss=1.4191, Noise=0.9383\n",
      "Iter 190: Loss=1.4184, Noise=0.9368\n",
      "Iter 191: Loss=1.4185, Noise=0.9352\n",
      "Iter 192: Loss=1.4181, Noise=0.9335\n",
      "Iter 193: Loss=1.4181, Noise=0.9321\n",
      "Iter 194: Loss=1.4180, Noise=0.9305\n",
      "Iter 195: Loss=1.4175, Noise=0.9291\n",
      "Iter 196: Loss=1.4161, Noise=0.9277\n",
      "Iter 197: Loss=1.4164, Noise=0.9263\n",
      "Iter 198: Loss=1.4159, Noise=0.9248\n",
      "Iter 199: Loss=1.4153, Noise=0.9232\n",
      "Iter 200: Loss=1.4153, Noise=0.9215\n",
      "Iter 201: Loss=1.4146, Noise=0.9196\n",
      "Iter 202: Loss=1.4146, Noise=0.9179\n",
      "Iter 203: Loss=1.4142, Noise=0.9159\n",
      "Iter 204: Loss=1.4135, Noise=0.9139\n",
      "Iter 205: Loss=1.4135, Noise=0.9122\n",
      "Iter 206: Loss=1.4121, Noise=0.9105\n",
      "Iter 207: Loss=1.4119, Noise=0.9086\n",
      "Iter 208: Loss=1.4120, Noise=0.9067\n",
      "Iter 209: Loss=1.4107, Noise=0.9045\n",
      "Iter 210: Loss=1.4103, Noise=0.9026\n",
      "Iter 211: Loss=1.4094, Noise=0.9010\n",
      "Iter 212: Loss=1.4093, Noise=0.8994\n",
      "Iter 213: Loss=1.4084, Noise=0.8978\n",
      "Iter 214: Loss=1.4077, Noise=0.8961\n",
      "Iter 215: Loss=1.4063, Noise=0.8945\n",
      "Iter 216: Loss=1.4066, Noise=0.8931\n",
      "Iter 217: Loss=1.4058, Noise=0.8915\n",
      "Iter 218: Loss=1.4056, Noise=0.8902\n",
      "Iter 219: Loss=1.4058, Noise=0.8885\n",
      "Iter 220: Loss=1.4049, Noise=0.8871\n",
      "Iter 221: Loss=1.4030, Noise=0.8855\n",
      "Iter 222: Loss=1.4033, Noise=0.8837\n",
      "Iter 223: Loss=1.4027, Noise=0.8821\n",
      "Iter 224: Loss=1.4022, Noise=0.8803\n",
      "Iter 225: Loss=1.4016, Noise=0.8786\n",
      "Iter 226: Loss=1.4003, Noise=0.8767\n",
      "Iter 227: Loss=1.3995, Noise=0.8747\n",
      "Iter 228: Loss=1.3990, Noise=0.8725\n",
      "Iter 229: Loss=1.3991, Noise=0.8701\n",
      "Iter 230: Loss=1.3981, Noise=0.8674\n",
      "Iter 231: Loss=1.3969, Noise=0.8644\n",
      "Iter 232: Loss=1.3968, Noise=0.8614\n",
      "Iter 233: Loss=1.3964, Noise=0.8584\n",
      "Iter 234: Loss=1.3953, Noise=0.8554\n",
      "Iter 235: Loss=1.3925, Noise=0.8520\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 236: Loss=1.3931, Noise=0.8487\n",
      "Iter 237: Loss=1.3912, Noise=0.8453\n",
      "Iter 238: Loss=1.3921, Noise=0.8418\n",
      "Iter 239: Loss=1.3904, Noise=0.8382\n",
      "Iter 240: Loss=1.3860, Noise=0.8345\n",
      "Iter 241: Loss=1.3864, Noise=0.8307\n",
      "Iter 242: Loss=1.3869, Noise=0.8270\n",
      "Iter 243: Loss=1.3845, Noise=0.8234\n",
      "Iter 244: Loss=1.3837, Noise=0.8197\n",
      "Iter 245: Loss=1.3827, Noise=0.8161\n",
      "Iter 246: Loss=1.3823, Noise=0.8125\n",
      "Iter 247: Loss=1.3819, Noise=0.8089\n",
      "Iter 248: Loss=1.3817, Noise=0.8054\n",
      "Iter 249: Loss=1.3793, Noise=0.8018\n",
      "Iter 250: Loss=1.3783, Noise=0.7981\n",
      "Iter 251: Loss=1.3770, Noise=0.7944\n",
      "Iter 252: Loss=1.3747, Noise=0.7907\n",
      "Iter 253: Loss=1.3742, Noise=0.7869\n",
      "Iter 254: Loss=1.3741, Noise=0.7830\n",
      "Iter 255: Loss=1.3733, Noise=0.7789\n",
      "Iter 256: Loss=1.3707, Noise=0.7748\n",
      "Iter 257: Loss=1.3703, Noise=0.7705\n",
      "Iter 258: Loss=1.3695, Noise=0.7661\n",
      "Iter 259: Loss=1.3665, Noise=0.7618\n",
      "Iter 260: Loss=1.3660, Noise=0.7575\n",
      "Iter 261: Loss=1.3662, Noise=0.7532\n",
      "Iter 262: Loss=1.3617, Noise=0.7489\n",
      "Iter 263: Loss=1.3624, Noise=0.7446\n",
      "Iter 264: Loss=1.3605, Noise=0.7401\n",
      "Iter 265: Loss=1.3591, Noise=0.7358\n",
      "Iter 266: Loss=1.3574, Noise=0.7315\n",
      "Iter 267: Loss=1.3539, Noise=0.7270\n",
      "Iter 268: Loss=1.3536, Noise=0.7224\n",
      "Iter 269: Loss=1.3550, Noise=0.7178\n",
      "Iter 270: Loss=1.3501, Noise=0.7132\n",
      "Iter 271: Loss=1.3495, Noise=0.7086\n",
      "Iter 272: Loss=1.3475, Noise=0.7039\n",
      "Iter 273: Loss=1.3442, Noise=0.6990\n",
      "Iter 274: Loss=1.3457, Noise=0.6940\n",
      "Iter 275: Loss=1.3442, Noise=0.6890\n",
      "Iter 276: Loss=1.3401, Noise=0.6838\n",
      "Iter 277: Loss=1.3388, Noise=0.6786\n",
      "Iter 278: Loss=1.3363, Noise=0.6734\n",
      "Iter 279: Loss=1.3333, Noise=0.6680\n",
      "Iter 280: Loss=1.3333, Noise=0.6625\n",
      "Iter 281: Loss=1.3325, Noise=0.6571\n",
      "Iter 282: Loss=1.3298, Noise=0.6517\n",
      "Iter 283: Loss=1.3282, Noise=0.6464\n",
      "Iter 284: Loss=1.3267, Noise=0.6410\n",
      "Iter 285: Loss=1.3215, Noise=0.6356\n",
      "Iter 286: Loss=1.3212, Noise=0.6303\n",
      "Iter 287: Loss=1.3185, Noise=0.6249\n",
      "Iter 288: Loss=1.3187, Noise=0.6198\n",
      "Iter 289: Loss=1.3144, Noise=0.6145\n",
      "Iter 290: Loss=1.3130, Noise=0.6092\n",
      "Iter 291: Loss=1.3139, Noise=0.6037\n",
      "Iter 292: Loss=1.3130, Noise=0.5983\n",
      "Iter 293: Loss=1.3098, Noise=0.5928\n",
      "Iter 294: Loss=1.3097, Noise=0.5876\n",
      "Iter 295: Loss=1.3093, Noise=0.5821\n",
      "Iter 296: Loss=1.2980, Noise=0.5767\n",
      "Iter 297: Loss=1.2941, Noise=0.5713\n",
      "Iter 298: Loss=1.2988, Noise=0.5658\n",
      "Iter 299: Loss=1.2948, Noise=0.5602\n",
      "Iter 300: Loss=1.2920, Noise=0.5547\n",
      "Iter 301: Loss=1.2893, Noise=0.5492\n",
      "Iter 302: Loss=1.2862, Noise=0.5436\n",
      "Iter 303: Loss=1.2855, Noise=0.5381\n",
      "Iter 304: Loss=1.2802, Noise=0.5325\n",
      "Iter 305: Loss=1.2781, Noise=0.5269\n",
      "Iter 306: Loss=1.2769, Noise=0.5212\n",
      "Iter 307: Loss=1.2769, Noise=0.5156\n",
      "Iter 308: Loss=1.2700, Noise=0.5099\n",
      "Iter 309: Loss=1.2702, Noise=0.5040\n",
      "Iter 310: Loss=1.2677, Noise=0.4982\n",
      "Iter 311: Loss=1.2662, Noise=0.4924\n",
      "Iter 312: Loss=1.2619, Noise=0.4865\n",
      "Iter 313: Loss=1.2604, Noise=0.4808\n",
      "Iter 314: Loss=1.2562, Noise=0.4750\n",
      "Iter 315: Loss=1.2562, Noise=0.4692\n",
      "Iter 316: Loss=1.2460, Noise=0.4635\n",
      "Iter 317: Loss=1.2469, Noise=0.4578\n",
      "Iter 318: Loss=1.2452, Noise=0.4520\n",
      "Iter 319: Loss=1.2376, Noise=0.4463\n",
      "Iter 320: Loss=1.2391, Noise=0.4405\n",
      "Iter 321: Loss=1.2415, Noise=0.4347\n",
      "Iter 322: Loss=1.2329, Noise=0.4290\n",
      "Iter 323: Loss=1.2261, Noise=0.4231\n",
      "Iter 324: Loss=1.2252, Noise=0.4173\n",
      "Iter 325: Loss=1.2162, Noise=0.4115\n",
      "Iter 326: Loss=1.2155, Noise=0.4057\n",
      "Iter 327: Loss=1.2163, Noise=0.3999\n",
      "Iter 328: Loss=1.2087, Noise=0.3941\n",
      "Iter 329: Loss=1.1976, Noise=0.3883\n",
      "Iter 330: Loss=1.2006, Noise=0.3827\n",
      "Iter 331: Loss=1.2083, Noise=0.3773\n",
      "Iter 332: Loss=1.1924, Noise=0.3719\n",
      "Iter 333: Loss=1.1937, Noise=0.3666\n",
      "Iter 334: Loss=1.1873, Noise=0.3613\n",
      "Iter 335: Loss=1.1830, Noise=0.3559\n",
      "Iter 336: Loss=1.1854, Noise=0.3506\n",
      "Iter 337: Loss=1.1844, Noise=0.3452\n",
      "Iter 338: Loss=1.1753, Noise=0.3399\n",
      "Iter 339: Loss=1.1763, Noise=0.3346\n",
      "Iter 340: Loss=1.1677, Noise=0.3294\n",
      "Iter 341: Loss=1.1661, Noise=0.3242\n",
      "Iter 342: Loss=1.1652, Noise=0.3190\n",
      "Iter 343: Loss=1.1510, Noise=0.3140\n",
      "Iter 344: Loss=1.1463, Noise=0.3089\n",
      "Iter 345: Loss=1.1407, Noise=0.3039\n",
      "Iter 346: Loss=1.1493, Noise=0.2989\n",
      "Iter 347: Loss=1.1361, Noise=0.2940\n",
      "Iter 348: Loss=1.1375, Noise=0.2891\n",
      "Iter 349: Loss=1.1280, Noise=0.2843\n",
      "Iter 350: Loss=1.1296, Noise=0.2795\n",
      "Iter 351: Loss=1.1153, Noise=0.2748\n",
      "Iter 352: Loss=1.1194, Noise=0.2702\n",
      "Iter 353: Loss=1.1072, Noise=0.2656\n",
      "Iter 354: Loss=1.1055, Noise=0.2611\n",
      "Iter 355: Loss=1.0956, Noise=0.2567\n",
      "Iter 356: Loss=1.1080, Noise=0.2523\n",
      "Iter 357: Loss=1.1006, Noise=0.2480\n",
      "Iter 358: Loss=1.0941, Noise=0.2438\n",
      "Iter 359: Loss=1.0891, Noise=0.2397\n",
      "Iter 360: Loss=1.0759, Noise=0.2355\n",
      "Iter 361: Loss=1.0712, Noise=0.2314\n",
      "Iter 362: Loss=1.0635, Noise=0.2273\n",
      "Iter 363: Loss=1.0680, Noise=0.2233\n",
      "Iter 364: Loss=1.0632, Noise=0.2194\n",
      "Iter 365: Loss=1.0508, Noise=0.2154\n",
      "Iter 366: Loss=1.0616, Noise=0.2116\n",
      "Iter 367: Loss=1.0518, Noise=0.2078\n",
      "Iter 368: Loss=1.0522, Noise=0.2041\n",
      "Iter 369: Loss=1.0391, Noise=0.2005\n",
      "Iter 370: Loss=1.0355, Noise=0.1969\n",
      "Iter 371: Loss=1.0406, Noise=0.1934\n",
      "Iter 372: Loss=1.0287, Noise=0.1899\n",
      "Iter 373: Loss=1.0249, Noise=0.1865\n",
      "Iter 374: Loss=1.0268, Noise=0.1832\n",
      "Iter 375: Loss=1.0214, Noise=0.1798\n",
      "Iter 376: Loss=1.0049, Noise=0.1766\n",
      "Iter 377: Loss=1.0105, Noise=0.1733\n",
      "Iter 378: Loss=1.0078, Noise=0.1702\n",
      "Iter 379: Loss=0.9931, Noise=0.1671\n",
      "Iter 380: Loss=0.9985, Noise=0.1641\n",
      "Iter 381: Loss=0.9948, Noise=0.1611\n",
      "Iter 382: Loss=0.9880, Noise=0.1583\n",
      "Iter 383: Loss=0.9743, Noise=0.1555\n",
      "Iter 384: Loss=0.9780, Noise=0.1527\n",
      "Iter 385: Loss=0.9622, Noise=0.1500\n",
      "Iter 386: Loss=0.9781, Noise=0.1474\n",
      "Iter 387: Loss=0.9586, Noise=0.1448\n",
      "Iter 388: Loss=0.9559, Noise=0.1423\n",
      "Iter 389: Loss=0.9497, Noise=0.1398\n",
      "Iter 390: Loss=0.9381, Noise=0.1373\n",
      "Iter 391: Loss=0.9434, Noise=0.1349\n",
      "Iter 392: Loss=0.9271, Noise=0.1325\n",
      "Iter 393: Loss=0.9445, Noise=0.1302\n",
      "Iter 394: Loss=0.9268, Noise=0.1279\n",
      "Iter 395: Loss=0.9370, Noise=0.1257\n",
      "Iter 396: Loss=0.8937, Noise=0.1235\n",
      "Iter 397: Loss=0.9020, Noise=0.1213\n",
      "Iter 398: Loss=0.9179, Noise=0.1192\n",
      "Iter 399: Loss=0.9120, Noise=0.1171\n",
      "Iter 400: Loss=0.9041, Noise=0.1151\n",
      "Iter 401: Loss=0.9026, Noise=0.1131\n",
      "Iter 402: Loss=0.8925, Noise=0.1113\n",
      "Iter 403: Loss=0.8891, Noise=0.1094\n",
      "Iter 404: Loss=0.8930, Noise=0.1075\n",
      "Iter 405: Loss=0.8985, Noise=0.1057\n",
      "Iter 406: Loss=0.8851, Noise=0.1040\n",
      "Iter 407: Loss=0.8767, Noise=0.1023\n",
      "Iter 408: Loss=0.8588, Noise=0.1006\n",
      "Iter 409: Loss=0.8869, Noise=0.0989\n",
      "Iter 410: Loss=0.8697, Noise=0.0973\n",
      "Iter 411: Loss=0.8715, Noise=0.0958\n",
      "Iter 412: Loss=0.8694, Noise=0.0943\n",
      "Iter 413: Loss=0.8728, Noise=0.0928\n",
      "Iter 414: Loss=0.8618, Noise=0.0913\n",
      "Iter 415: Loss=0.8589, Noise=0.0899\n",
      "Iter 416: Loss=0.8375, Noise=0.0885\n",
      "Iter 417: Loss=0.8532, Noise=0.0871\n",
      "Iter 418: Loss=0.8425, Noise=0.0858\n",
      "Iter 419: Loss=0.8612, Noise=0.0845\n",
      "Iter 420: Loss=0.8295, Noise=0.0832\n",
      "Iter 421: Loss=0.8365, Noise=0.0819\n",
      "Iter 422: Loss=0.8241, Noise=0.0807\n",
      "Iter 423: Loss=0.8202, Noise=0.0795\n",
      "Iter 424: Loss=0.8228, Noise=0.0783\n",
      "Iter 425: Loss=0.8154, Noise=0.0772\n",
      "Iter 426: Loss=0.8254, Noise=0.0761\n",
      "Iter 427: Loss=0.8206, Noise=0.0749\n",
      "Iter 428: Loss=0.8258, Noise=0.0739\n",
      "Iter 429: Loss=0.7996, Noise=0.0728\n",
      "Iter 430: Loss=0.7787, Noise=0.0718\n",
      "Iter 431: Loss=0.7829, Noise=0.0707\n",
      "Iter 432: Loss=0.7914, Noise=0.0697\n",
      "Iter 433: Loss=0.7667, Noise=0.0687\n",
      "Iter 434: Loss=0.7848, Noise=0.0677\n",
      "Iter 435: Loss=0.7858, Noise=0.0667\n",
      "Iter 436: Loss=0.7836, Noise=0.0658\n",
      "Iter 437: Loss=0.7624, Noise=0.0648\n",
      "Iter 438: Loss=0.7901, Noise=0.0639\n",
      "Iter 439: Loss=0.7672, Noise=0.0630\n",
      "Iter 440: Loss=0.7525, Noise=0.0622\n",
      "Iter 441: Loss=0.7816, Noise=0.0613\n",
      "Iter 442: Loss=0.7456, Noise=0.0605\n",
      "Iter 443: Loss=0.7449, Noise=0.0597\n",
      "Iter 444: Loss=0.7536, Noise=0.0588\n",
      "Iter 445: Loss=0.7666, Noise=0.0580\n",
      "Iter 446: Loss=0.7703, Noise=0.0572\n",
      "Iter 447: Loss=0.7619, Noise=0.0565\n",
      "Iter 448: Loss=0.7507, Noise=0.0557\n",
      "Iter 449: Loss=0.7313, Noise=0.0550\n",
      "Iter 450: Loss=0.7354, Noise=0.0543\n",
      "Iter 451: Loss=0.7093, Noise=0.0536\n",
      "Iter 452: Loss=0.7373, Noise=0.0529\n",
      "Iter 453: Loss=0.7478, Noise=0.0522\n",
      "Iter 454: Loss=0.7104, Noise=0.0515\n",
      "Iter 455: Loss=0.7453, Noise=0.0509\n",
      "Iter 456: Loss=0.7184, Noise=0.0502\n",
      "Iter 457: Loss=0.7093, Noise=0.0496\n",
      "Iter 458: Loss=0.7055, Noise=0.0490\n",
      "Iter 459: Loss=0.6889, Noise=0.0484\n",
      "Iter 460: Loss=0.7166, Noise=0.0478\n",
      "Iter 461: Loss=0.7048, Noise=0.0472\n",
      "Iter 462: Loss=0.7131, Noise=0.0466\n",
      "Iter 463: Loss=0.6953, Noise=0.0460\n",
      "Iter 464: Loss=0.7253, Noise=0.0455\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 465: Loss=0.7192, Noise=0.0449\n",
      "Iter 466: Loss=0.6982, Noise=0.0444\n",
      "Iter 467: Loss=0.6868, Noise=0.0438\n",
      "Iter 468: Loss=0.6968, Noise=0.0433\n",
      "Iter 469: Loss=0.7079, Noise=0.0428\n",
      "Iter 470: Loss=0.6471, Noise=0.0423\n",
      "Iter 471: Loss=0.6836, Noise=0.0418\n",
      "Iter 472: Loss=0.7049, Noise=0.0413\n",
      "Iter 473: Loss=0.6410, Noise=0.0408\n",
      "Iter 474: Loss=0.6784, Noise=0.0403\n",
      "Iter 475: Loss=0.6943, Noise=0.0398\n",
      "Iter 476: Loss=0.6646, Noise=0.0394\n",
      "Iter 477: Loss=0.6339, Noise=0.0389\n",
      "Iter 478: Loss=0.6752, Noise=0.0385\n",
      "Iter 479: Loss=0.6788, Noise=0.0380\n",
      "Iter 480: Loss=0.6647, Noise=0.0376\n",
      "Iter 481: Loss=0.6417, Noise=0.0371\n",
      "Iter 482: Loss=0.6307, Noise=0.0367\n",
      "Iter 483: Loss=0.6724, Noise=0.0363\n",
      "Iter 484: Loss=0.6697, Noise=0.0359\n",
      "Iter 485: Loss=0.6904, Noise=0.0355\n",
      "Iter 486: Loss=0.6433, Noise=0.0351\n",
      "Iter 487: Loss=0.6461, Noise=0.0347\n",
      "Iter 488: Loss=0.6522, Noise=0.0343\n",
      "Iter 489: Loss=0.6363, Noise=0.0340\n",
      "Iter 490: Loss=0.6407, Noise=0.0336\n",
      "Iter 491: Loss=0.6291, Noise=0.0332\n",
      "Iter 492: Loss=0.6267, Noise=0.0329\n",
      "Iter 493: Loss=0.6200, Noise=0.0325\n",
      "Iter 494: Loss=0.6468, Noise=0.0322\n",
      "Iter 495: Loss=0.6053, Noise=0.0318\n",
      "Iter 496: Loss=0.6335, Noise=0.0315\n",
      "Iter 497: Loss=0.6345, Noise=0.0312\n",
      "Iter 498: Loss=0.6487, Noise=0.0309\n",
      "Iter 499: Loss=0.6062, Noise=0.0305\n",
      "Iter 500: Loss=0.6292, Noise=0.0302\n",
      "Iter 501: Loss=0.6371, Noise=0.0299\n",
      "Iter 502: Loss=0.6355, Noise=0.0296\n",
      "Iter 503: Loss=0.6257, Noise=0.0294\n",
      "Iter 504: Loss=0.6115, Noise=0.0291\n",
      "Iter 505: Loss=0.6260, Noise=0.0288\n",
      "Iter 506: Loss=0.6132, Noise=0.0285\n",
      "Iter 507: Loss=0.6316, Noise=0.0282\n",
      "Iter 508: Loss=0.6106, Noise=0.0280\n",
      "Iter 509: Loss=0.5918, Noise=0.0277\n",
      "Iter 510: Loss=0.6014, Noise=0.0274\n",
      "Iter 511: Loss=0.6175, Noise=0.0271\n",
      "Iter 512: Loss=0.6098, Noise=0.0269\n",
      "Iter 513: Loss=0.5912, Noise=0.0266\n",
      "Iter 514: Loss=0.5911, Noise=0.0264\n",
      "Iter 515: Loss=0.5901, Noise=0.0261\n",
      "Iter 516: Loss=0.5937, Noise=0.0259\n",
      "Iter 517: Loss=0.6170, Noise=0.0256\n",
      "Iter 518: Loss=0.5853, Noise=0.0254\n",
      "Iter 519: Loss=0.6088, Noise=0.0252\n",
      "Iter 520: Loss=0.5565, Noise=0.0249\n",
      "Iter 521: Loss=0.5768, Noise=0.0247\n",
      "Iter 522: Loss=0.6044, Noise=0.0245\n",
      "Iter 523: Loss=0.5865, Noise=0.0243\n",
      "Iter 524: Loss=0.5834, Noise=0.0241\n",
      "Iter 525: Loss=0.5718, Noise=0.0238\n",
      "Iter 526: Loss=0.5478, Noise=0.0236\n",
      "Iter 527: Loss=0.5572, Noise=0.0234\n",
      "Iter 528: Loss=0.5871, Noise=0.0232\n",
      "Iter 529: Loss=0.5518, Noise=0.0230\n",
      "Iter 530: Loss=0.5712, Noise=0.0228\n",
      "Iter 531: Loss=0.5739, Noise=0.0226\n",
      "Iter 532: Loss=0.5766, Noise=0.0224\n",
      "Iter 533: Loss=0.5912, Noise=0.0222\n",
      "Iter 534: Loss=0.5779, Noise=0.0221\n",
      "Iter 535: Loss=0.5746, Noise=0.0219\n",
      "Iter 536: Loss=0.5495, Noise=0.0217\n",
      "Iter 537: Loss=0.5584, Noise=0.0215\n",
      "Iter 538: Loss=0.5810, Noise=0.0213\n",
      "Iter 539: Loss=0.5437, Noise=0.0212\n",
      "Iter 540: Loss=0.5543, Noise=0.0210\n",
      "Iter 541: Loss=0.5489, Noise=0.0208\n",
      "Iter 542: Loss=0.5472, Noise=0.0206\n",
      "Iter 543: Loss=0.5754, Noise=0.0205\n",
      "Iter 544: Loss=0.5416, Noise=0.0203\n",
      "Iter 545: Loss=0.5440, Noise=0.0201\n",
      "Iter 546: Loss=0.5406, Noise=0.0200\n",
      "Iter 547: Loss=0.5484, Noise=0.0198\n",
      "Iter 548: Loss=0.5537, Noise=0.0197\n",
      "Iter 549: Loss=0.5472, Noise=0.0195\n",
      "Iter 550: Loss=0.5043, Noise=0.0194\n",
      "Iter 551: Loss=0.5176, Noise=0.0192\n",
      "Iter 552: Loss=0.5538, Noise=0.0190\n",
      "Iter 553: Loss=0.4980, Noise=0.0189\n",
      "Iter 554: Loss=0.5623, Noise=0.0187\n",
      "Iter 555: Loss=0.5392, Noise=0.0186\n",
      "Iter 556: Loss=0.5243, Noise=0.0185\n",
      "Iter 557: Loss=0.5284, Noise=0.0183\n",
      "Iter 558: Loss=0.5548, Noise=0.0182\n",
      "Iter 559: Loss=0.5220, Noise=0.0180\n",
      "Iter 560: Loss=0.5182, Noise=0.0179\n",
      "Iter 561: Loss=0.5118, Noise=0.0178\n",
      "Iter 562: Loss=0.5037, Noise=0.0176\n",
      "Iter 563: Loss=0.5343, Noise=0.0175\n",
      "Iter 564: Loss=0.5415, Noise=0.0174\n",
      "Iter 565: Loss=0.5268, Noise=0.0173\n",
      "Iter 566: Loss=0.5102, Noise=0.0171\n",
      "Iter 567: Loss=0.5332, Noise=0.0170\n",
      "Iter 568: Loss=0.5132, Noise=0.0169\n",
      "Iter 569: Loss=0.4862, Noise=0.0168\n",
      "Iter 570: Loss=0.5328, Noise=0.0167\n",
      "Iter 571: Loss=0.5079, Noise=0.0165\n",
      "Iter 572: Loss=0.5277, Noise=0.0164\n",
      "Iter 573: Loss=0.4775, Noise=0.0163\n",
      "Iter 574: Loss=0.4991, Noise=0.0162\n",
      "Iter 575: Loss=0.5198, Noise=0.0161\n",
      "Iter 576: Loss=0.4846, Noise=0.0160\n",
      "Iter 577: Loss=0.5113, Noise=0.0159\n",
      "Iter 578: Loss=0.4989, Noise=0.0158\n",
      "Iter 579: Loss=0.5023, Noise=0.0157\n",
      "Iter 580: Loss=0.5147, Noise=0.0156\n",
      "Iter 581: Loss=0.5385, Noise=0.0155\n",
      "Iter 582: Loss=0.5301, Noise=0.0154\n",
      "Iter 583: Loss=0.5057, Noise=0.0153\n",
      "Iter 584: Loss=0.4732, Noise=0.0152\n",
      "Iter 585: Loss=0.4964, Noise=0.0151\n",
      "Iter 586: Loss=0.4609, Noise=0.0150\n",
      "Iter 587: Loss=0.4971, Noise=0.0149\n",
      "Iter 588: Loss=0.4749, Noise=0.0148\n",
      "Iter 589: Loss=0.4730, Noise=0.0147\n",
      "Iter 590: Loss=0.5263, Noise=0.0146\n",
      "Iter 591: Loss=0.5100, Noise=0.0145\n",
      "Iter 592: Loss=0.4613, Noise=0.0145\n",
      "Iter 593: Loss=0.4718, Noise=0.0144\n",
      "Iter 594: Loss=0.4899, Noise=0.0143\n",
      "Iter 595: Loss=0.5030, Noise=0.0142\n",
      "Iter 596: Loss=0.4838, Noise=0.0141\n",
      "Iter 597: Loss=0.5083, Noise=0.0140\n",
      "Iter 598: Loss=0.4653, Noise=0.0140\n",
      "Iter 599: Loss=0.4787, Noise=0.0139\n",
      "Iter 600: Loss=0.4688, Noise=0.0138\n",
      "Iter 601: Loss=0.4753, Noise=0.0137\n",
      "Iter 602: Loss=0.4742, Noise=0.0137\n",
      "Iter 603: Loss=0.4885, Noise=0.0136\n",
      "Iter 604: Loss=0.4414, Noise=0.0135\n",
      "Iter 605: Loss=0.4863, Noise=0.0134\n",
      "Iter 606: Loss=0.4920, Noise=0.0134\n",
      "Iter 607: Loss=0.4814, Noise=0.0133\n",
      "Iter 608: Loss=0.4457, Noise=0.0132\n",
      "Iter 609: Loss=0.4412, Noise=0.0131\n",
      "Iter 610: Loss=0.4765, Noise=0.0131\n",
      "Iter 611: Loss=0.4789, Noise=0.0130\n",
      "Iter 612: Loss=0.4694, Noise=0.0129\n",
      "Iter 613: Loss=0.4882, Noise=0.0129\n",
      "Iter 614: Loss=0.4715, Noise=0.0128\n",
      "Iter 615: Loss=0.4815, Noise=0.0127\n",
      "Iter 616: Loss=0.4560, Noise=0.0127\n",
      "Iter 617: Loss=0.4606, Noise=0.0126\n",
      "Iter 618: Loss=0.4409, Noise=0.0126\n",
      "Iter 619: Loss=0.4505, Noise=0.0125\n",
      "Iter 620: Loss=0.4756, Noise=0.0124\n",
      "Iter 621: Loss=0.4559, Noise=0.0124\n",
      "Iter 622: Loss=0.4689, Noise=0.0123\n",
      "Iter 623: Loss=0.4455, Noise=0.0123\n",
      "Iter 624: Loss=0.4673, Noise=0.0122\n",
      "Iter 625: Loss=0.4174, Noise=0.0121\n",
      "Iter 626: Loss=0.4225, Noise=0.0121\n",
      "Iter 627: Loss=0.4630, Noise=0.0120\n",
      "Iter 628: Loss=0.4237, Noise=0.0120\n",
      "Iter 629: Loss=0.4227, Noise=0.0119\n",
      "Iter 630: Loss=0.4588, Noise=0.0118\n",
      "Iter 631: Loss=0.4014, Noise=0.0118\n",
      "Iter 632: Loss=0.4356, Noise=0.0117\n",
      "Iter 633: Loss=0.4853, Noise=0.0117\n",
      "Iter 634: Loss=0.4427, Noise=0.0116\n",
      "Iter 635: Loss=0.4425, Noise=0.0116\n",
      "Iter 636: Loss=0.4524, Noise=0.0115\n",
      "Iter 637: Loss=0.4475, Noise=0.0115\n",
      "Iter 638: Loss=0.4192, Noise=0.0114\n",
      "Iter 639: Loss=0.4086, Noise=0.0114\n",
      "Iter 640: Loss=0.4266, Noise=0.0113\n",
      "Iter 641: Loss=0.4733, Noise=0.0113\n",
      "Iter 642: Loss=0.4502, Noise=0.0112\n",
      "Iter 643: Loss=0.4201, Noise=0.0112\n",
      "Iter 644: Loss=0.4578, Noise=0.0111\n",
      "Iter 645: Loss=0.4045, Noise=0.0111\n",
      "Iter 646: Loss=0.4494, Noise=0.0110\n",
      "Iter 647: Loss=0.4630, Noise=0.0110\n",
      "Iter 648: Loss=0.4217, Noise=0.0109\n",
      "Iter 649: Loss=0.4183, Noise=0.0109\n",
      "Iter 650: Loss=0.4020, Noise=0.0108\n",
      "Iter 651: Loss=0.4438, Noise=0.0108\n",
      "Iter 652: Loss=0.4311, Noise=0.0107\n",
      "Iter 653: Loss=0.4186, Noise=0.0107\n",
      "Iter 654: Loss=0.4079, Noise=0.0107\n",
      "Iter 655: Loss=0.3840, Noise=0.0106\n",
      "Iter 656: Loss=0.4245, Noise=0.0106\n",
      "Iter 657: Loss=0.4190, Noise=0.0105\n",
      "Iter 658: Loss=0.4741, Noise=0.0105\n",
      "Iter 659: Loss=0.3886, Noise=0.0104\n",
      "Iter 660: Loss=0.4345, Noise=0.0104\n",
      "Iter 661: Loss=0.4523, Noise=0.0104\n",
      "Iter 662: Loss=0.3912, Noise=0.0103\n",
      "Iter 663: Loss=0.4117, Noise=0.0103\n",
      "Iter 664: Loss=0.4232, Noise=0.0103\n",
      "Iter 665: Loss=0.4133, Noise=0.0102\n",
      "Iter 666: Loss=0.4013, Noise=0.0102\n",
      "Iter 667: Loss=0.4230, Noise=0.0101\n",
      "Iter 668: Loss=0.4202, Noise=0.0101\n",
      "Iter 669: Loss=0.3847, Noise=0.0101\n",
      "Iter 670: Loss=0.4256, Noise=0.0100\n",
      "Iter 671: Loss=0.4155, Noise=0.0100\n",
      "Iter 672: Loss=0.3668, Noise=0.0100\n",
      "Iter 673: Loss=0.4110, Noise=0.0099\n",
      "Iter 674: Loss=0.4107, Noise=0.0099\n",
      "Iter 675: Loss=0.3973, Noise=0.0099\n",
      "Iter 676: Loss=0.4064, Noise=0.0098\n",
      "Iter 677: Loss=0.4329, Noise=0.0098\n",
      "Iter 678: Loss=0.4148, Noise=0.0098\n",
      "Iter 679: Loss=0.4125, Noise=0.0097\n",
      "Iter 680: Loss=0.3883, Noise=0.0097\n",
      "Iter 681: Loss=0.4121, Noise=0.0097\n",
      "Iter 682: Loss=0.4179, Noise=0.0096\n",
      "Iter 683: Loss=0.4485, Noise=0.0096\n",
      "Iter 684: Loss=0.4159, Noise=0.0096\n",
      "Iter 685: Loss=0.4319, Noise=0.0096\n",
      "Iter 686: Loss=0.3943, Noise=0.0095\n",
      "Iter 687: Loss=0.4362, Noise=0.0095\n",
      "Iter 688: Loss=0.3719, Noise=0.0095\n",
      "Iter 689: Loss=0.3984, Noise=0.0094\n",
      "Iter 690: Loss=0.4261, Noise=0.0094\n",
      "Iter 691: Loss=0.4059, Noise=0.0094\n",
      "Iter 692: Loss=0.3949, Noise=0.0094\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 693: Loss=0.3746, Noise=0.0093\n",
      "Iter 694: Loss=0.3958, Noise=0.0093\n",
      "Iter 695: Loss=0.4249, Noise=0.0093\n",
      "Iter 696: Loss=0.3949, Noise=0.0092\n",
      "Iter 697: Loss=0.3949, Noise=0.0092\n",
      "Iter 698: Loss=0.3866, Noise=0.0092\n",
      "Iter 699: Loss=0.3870, Noise=0.0092\n",
      "Iter 700: Loss=0.3977, Noise=0.0091\n",
      "Iter 701: Loss=0.3624, Noise=0.0091\n",
      "Iter 702: Loss=0.3890, Noise=0.0091\n",
      "Iter 703: Loss=0.3981, Noise=0.0091\n",
      "Iter 704: Loss=0.3933, Noise=0.0090\n",
      "Iter 705: Loss=0.3681, Noise=0.0090\n",
      "Iter 706: Loss=0.3862, Noise=0.0090\n",
      "Iter 707: Loss=0.3928, Noise=0.0090\n",
      "Iter 708: Loss=0.3758, Noise=0.0089\n",
      "Iter 709: Loss=0.3571, Noise=0.0089\n",
      "Iter 710: Loss=0.3494, Noise=0.0089\n",
      "Iter 711: Loss=0.4096, Noise=0.0089\n",
      "Iter 712: Loss=0.3800, Noise=0.0088\n",
      "Iter 713: Loss=0.3856, Noise=0.0088\n",
      "Iter 714: Loss=0.3888, Noise=0.0088\n",
      "Iter 715: Loss=0.3622, Noise=0.0088\n",
      "Iter 716: Loss=0.3661, Noise=0.0088\n",
      "Iter 717: Loss=0.3882, Noise=0.0087\n",
      "Iter 718: Loss=0.3801, Noise=0.0087\n",
      "Iter 719: Loss=0.3620, Noise=0.0087\n",
      "Iter 720: Loss=0.3999, Noise=0.0087\n",
      "Iter 721: Loss=0.3721, Noise=0.0087\n",
      "Iter 722: Loss=0.3890, Noise=0.0086\n",
      "Iter 723: Loss=0.3976, Noise=0.0086\n",
      "Iter 724: Loss=0.3808, Noise=0.0086\n",
      "Iter 725: Loss=0.3900, Noise=0.0086\n",
      "Iter 726: Loss=0.3601, Noise=0.0086\n",
      "Iter 727: Loss=0.3887, Noise=0.0085\n",
      "Iter 728: Loss=0.3830, Noise=0.0085\n",
      "Iter 729: Loss=0.3987, Noise=0.0085\n",
      "Iter 730: Loss=0.3748, Noise=0.0085\n",
      "Iter 731: Loss=0.3679, Noise=0.0085\n",
      "Iter 732: Loss=0.3725, Noise=0.0084\n",
      "Iter 733: Loss=0.3542, Noise=0.0084\n",
      "Iter 734: Loss=0.3868, Noise=0.0084\n",
      "Iter 735: Loss=0.3416, Noise=0.0084\n",
      "Iter 736: Loss=0.3846, Noise=0.0084\n",
      "Iter 737: Loss=0.3750, Noise=0.0084\n",
      "Iter 738: Loss=0.3879, Noise=0.0083\n",
      "Iter 739: Loss=0.3902, Noise=0.0083\n",
      "Iter 740: Loss=0.3440, Noise=0.0083\n",
      "Iter 741: Loss=0.3544, Noise=0.0083\n",
      "Iter 742: Loss=0.3574, Noise=0.0083\n",
      "Iter 743: Loss=0.4130, Noise=0.0083\n",
      "Iter 744: Loss=0.3733, Noise=0.0083\n",
      "Iter 745: Loss=0.3846, Noise=0.0082\n",
      "Iter 746: Loss=0.3387, Noise=0.0082\n",
      "Iter 747: Loss=0.3421, Noise=0.0082\n",
      "Iter 748: Loss=0.3892, Noise=0.0082\n",
      "Iter 749: Loss=0.3492, Noise=0.0082\n",
      "Iter 750: Loss=0.3352, Noise=0.0082\n",
      "Iter 751: Loss=0.3452, Noise=0.0081\n",
      "Iter 752: Loss=0.3408, Noise=0.0081\n",
      "Iter 753: Loss=0.3678, Noise=0.0081\n",
      "Iter 754: Loss=0.3275, Noise=0.0081\n",
      "Iter 755: Loss=0.3598, Noise=0.0081\n",
      "Iter 756: Loss=0.3513, Noise=0.0081\n",
      "Iter 757: Loss=0.3736, Noise=0.0081\n",
      "Iter 758: Loss=0.3737, Noise=0.0080\n",
      "Iter 759: Loss=0.3516, Noise=0.0080\n",
      "Iter 760: Loss=0.3516, Noise=0.0080\n",
      "Iter 761: Loss=0.3459, Noise=0.0080\n",
      "Iter 762: Loss=0.3452, Noise=0.0080\n",
      "Iter 763: Loss=0.3406, Noise=0.0080\n",
      "Iter 764: Loss=0.3612, Noise=0.0080\n",
      "Iter 765: Loss=0.3798, Noise=0.0079\n",
      "Iter 766: Loss=0.3527, Noise=0.0079\n",
      "Iter 767: Loss=0.3707, Noise=0.0079\n",
      "Iter 768: Loss=0.3361, Noise=0.0079\n",
      "Iter 769: Loss=0.3712, Noise=0.0079\n",
      "Iter 770: Loss=0.3506, Noise=0.0079\n",
      "Iter 771: Loss=0.3363, Noise=0.0079\n",
      "Iter 772: Loss=0.3564, Noise=0.0079\n",
      "Iter 773: Loss=0.3551, Noise=0.0079\n",
      "Iter 774: Loss=0.3485, Noise=0.0078\n",
      "Iter 775: Loss=0.3893, Noise=0.0078\n",
      "Iter 776: Loss=0.3470, Noise=0.0078\n",
      "Iter 777: Loss=0.3466, Noise=0.0078\n",
      "Iter 778: Loss=0.3676, Noise=0.0078\n",
      "Iter 779: Loss=0.3484, Noise=0.0078\n",
      "Iter 780: Loss=0.3045, Noise=0.0078\n",
      "Iter 781: Loss=0.3389, Noise=0.0078\n",
      "Iter 782: Loss=0.3146, Noise=0.0078\n",
      "Iter 783: Loss=0.3651, Noise=0.0078\n",
      "Iter 784: Loss=0.3295, Noise=0.0077\n",
      "Iter 785: Loss=0.3242, Noise=0.0077\n",
      "Iter 786: Loss=0.3571, Noise=0.0077\n",
      "Iter 787: Loss=0.3412, Noise=0.0077\n",
      "Iter 788: Loss=0.3170, Noise=0.0077\n",
      "Iter 789: Loss=0.3115, Noise=0.0077\n",
      "Iter 790: Loss=0.3352, Noise=0.0077\n",
      "Iter 791: Loss=0.3468, Noise=0.0077\n",
      "Iter 792: Loss=0.3112, Noise=0.0076\n",
      "Iter 793: Loss=0.3328, Noise=0.0076\n",
      "Iter 794: Loss=0.3281, Noise=0.0076\n",
      "Iter 795: Loss=0.3443, Noise=0.0076\n",
      "Iter 796: Loss=0.3394, Noise=0.0076\n",
      "Iter 797: Loss=0.3487, Noise=0.0076\n",
      "Iter 798: Loss=0.3096, Noise=0.0076\n",
      "Iter 799: Loss=0.3397, Noise=0.0076\n",
      "Iter 800: Loss=0.3058, Noise=0.0076\n",
      "Iter 801: Loss=0.3098, Noise=0.0075\n",
      "Iter 802: Loss=0.3673, Noise=0.0075\n",
      "Iter 803: Loss=0.3375, Noise=0.0075\n",
      "Iter 804: Loss=0.3333, Noise=0.0075\n",
      "Iter 805: Loss=0.3219, Noise=0.0075\n",
      "Iter 806: Loss=0.3409, Noise=0.0075\n",
      "Iter 807: Loss=0.3261, Noise=0.0075\n",
      "Iter 808: Loss=0.2647, Noise=0.0075\n",
      "Iter 809: Loss=0.3346, Noise=0.0075\n",
      "Iter 810: Loss=0.3270, Noise=0.0075\n",
      "Iter 811: Loss=0.3422, Noise=0.0075\n",
      "Iter 812: Loss=0.3531, Noise=0.0074\n",
      "Iter 813: Loss=0.3557, Noise=0.0074\n",
      "Iter 814: Loss=0.3089, Noise=0.0074\n",
      "Iter 815: Loss=0.2859, Noise=0.0074\n",
      "Iter 816: Loss=0.2968, Noise=0.0074\n",
      "Iter 817: Loss=0.3107, Noise=0.0074\n",
      "Iter 818: Loss=0.3229, Noise=0.0074\n",
      "Iter 819: Loss=0.2859, Noise=0.0074\n",
      "Iter 820: Loss=0.3132, Noise=0.0074\n",
      "Iter 821: Loss=0.3164, Noise=0.0074\n",
      "Iter 822: Loss=0.3325, Noise=0.0074\n",
      "Iter 823: Loss=0.3341, Noise=0.0074\n",
      "Iter 824: Loss=0.3333, Noise=0.0074\n",
      "Iter 825: Loss=0.3559, Noise=0.0073\n",
      "Iter 826: Loss=0.2801, Noise=0.0073\n",
      "Iter 827: Loss=0.3343, Noise=0.0073\n",
      "Iter 828: Loss=0.3355, Noise=0.0073\n",
      "Iter 829: Loss=0.3553, Noise=0.0073\n",
      "Iter 830: Loss=0.3180, Noise=0.0073\n",
      "Iter 831: Loss=0.2930, Noise=0.0073\n",
      "Iter 832: Loss=0.2877, Noise=0.0073\n",
      "Iter 833: Loss=0.3099, Noise=0.0073\n",
      "Iter 834: Loss=0.3229, Noise=0.0073\n",
      "Iter 835: Loss=0.3751, Noise=0.0073\n",
      "Iter 836: Loss=0.3183, Noise=0.0073\n",
      "Iter 837: Loss=0.3147, Noise=0.0073\n",
      "Iter 838: Loss=0.3180, Noise=0.0073\n",
      "Iter 839: Loss=0.3479, Noise=0.0073\n",
      "Iter 840: Loss=0.3237, Noise=0.0073\n",
      "Iter 841: Loss=0.3032, Noise=0.0073\n",
      "Iter 842: Loss=0.3240, Noise=0.0073\n",
      "Iter 843: Loss=0.3086, Noise=0.0072\n",
      "Iter 844: Loss=0.3415, Noise=0.0072\n",
      "Iter 845: Loss=0.3457, Noise=0.0072\n",
      "Iter 846: Loss=0.2834, Noise=0.0072\n",
      "Iter 847: Loss=0.3318, Noise=0.0072\n",
      "Iter 848: Loss=0.3069, Noise=0.0072\n",
      "Iter 849: Loss=0.3535, Noise=0.0072\n",
      "Iter 850: Loss=0.3167, Noise=0.0072\n",
      "Iter 851: Loss=0.3053, Noise=0.0072\n",
      "Iter 852: Loss=0.2812, Noise=0.0072\n",
      "Iter 853: Loss=0.3304, Noise=0.0072\n",
      "Iter 854: Loss=0.3214, Noise=0.0072\n",
      "Iter 855: Loss=0.2845, Noise=0.0072\n",
      "Iter 856: Loss=0.2743, Noise=0.0072\n",
      "Iter 857: Loss=0.3290, Noise=0.0072\n",
      "Iter 858: Loss=0.3159, Noise=0.0072\n",
      "Iter 859: Loss=0.2862, Noise=0.0072\n",
      "Iter 860: Loss=0.3233, Noise=0.0072\n",
      "Iter 861: Loss=0.3110, Noise=0.0072\n",
      "Iter 862: Loss=0.3265, Noise=0.0072\n",
      "Iter 863: Loss=0.3146, Noise=0.0072\n",
      "Iter 864: Loss=0.3362, Noise=0.0072\n",
      "Iter 865: Loss=0.3285, Noise=0.0072\n",
      "Iter 866: Loss=0.3381, Noise=0.0072\n",
      "Iter 867: Loss=0.3212, Noise=0.0072\n",
      "Iter 868: Loss=0.3173, Noise=0.0071\n",
      "Iter 869: Loss=0.2968, Noise=0.0071\n",
      "Iter 870: Loss=0.3012, Noise=0.0071\n",
      "Iter 871: Loss=0.2723, Noise=0.0071\n",
      "Iter 872: Loss=0.2967, Noise=0.0071\n",
      "Iter 873: Loss=0.2965, Noise=0.0071\n",
      "Iter 874: Loss=0.2977, Noise=0.0071\n",
      "Iter 875: Loss=0.3215, Noise=0.0071\n",
      "Iter 876: Loss=0.2441, Noise=0.0071\n",
      "Iter 877: Loss=0.3464, Noise=0.0071\n",
      "Iter 878: Loss=0.3328, Noise=0.0071\n",
      "Iter 879: Loss=0.3368, Noise=0.0071\n",
      "Iter 880: Loss=0.2935, Noise=0.0071\n",
      "Iter 881: Loss=0.3241, Noise=0.0071\n",
      "Iter 882: Loss=0.3308, Noise=0.0071\n",
      "Iter 883: Loss=0.3154, Noise=0.0071\n",
      "Iter 884: Loss=0.3045, Noise=0.0071\n",
      "Iter 885: Loss=0.3147, Noise=0.0071\n",
      "Iter 886: Loss=0.3014, Noise=0.0071\n",
      "Iter 887: Loss=0.2773, Noise=0.0071\n",
      "Iter 888: Loss=0.3081, Noise=0.0071\n",
      "Iter 889: Loss=0.3051, Noise=0.0071\n",
      "Iter 890: Loss=0.3198, Noise=0.0071\n",
      "Iter 891: Loss=0.2640, Noise=0.0071\n",
      "Iter 892: Loss=0.3069, Noise=0.0070\n",
      "Iter 893: Loss=0.2810, Noise=0.0070\n",
      "Iter 894: Loss=0.2964, Noise=0.0070\n",
      "Iter 895: Loss=0.3015, Noise=0.0070\n",
      "Iter 896: Loss=0.3162, Noise=0.0070\n",
      "Iter 897: Loss=0.3346, Noise=0.0070\n",
      "Iter 898: Loss=0.3304, Noise=0.0070\n",
      "Iter 899: Loss=0.2954, Noise=0.0070\n",
      "Iter 900: Loss=0.2974, Noise=0.0070\n",
      "Iter 901: Loss=0.3302, Noise=0.0070\n",
      "Iter 902: Loss=0.2850, Noise=0.0070\n",
      "Iter 903: Loss=0.2566, Noise=0.0070\n",
      "Iter 904: Loss=0.2981, Noise=0.0070\n",
      "Iter 905: Loss=0.2785, Noise=0.0070\n",
      "Iter 906: Loss=0.2992, Noise=0.0070\n",
      "Iter 907: Loss=0.2669, Noise=0.0070\n",
      "Iter 908: Loss=0.3010, Noise=0.0070\n",
      "Iter 909: Loss=0.3045, Noise=0.0070\n",
      "Iter 910: Loss=0.2767, Noise=0.0070\n",
      "Iter 911: Loss=0.2818, Noise=0.0070\n",
      "Iter 912: Loss=0.2494, Noise=0.0070\n",
      "Iter 913: Loss=0.2693, Noise=0.0070\n",
      "Iter 914: Loss=0.3026, Noise=0.0070\n",
      "Iter 915: Loss=0.3035, Noise=0.0070\n",
      "Iter 916: Loss=0.3187, Noise=0.0069\n",
      "Iter 917: Loss=0.3060, Noise=0.0069\n",
      "Iter 918: Loss=0.2746, Noise=0.0069\n",
      "Iter 919: Loss=0.2915, Noise=0.0069\n",
      "Iter 920: Loss=0.3056, Noise=0.0069\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Iter 921: Loss=0.3006, Noise=0.0069\n",
      "Iter 922: Loss=0.2771, Noise=0.0069\n",
      "Iter 923: Loss=0.2909, Noise=0.0069\n",
      "Iter 924: Loss=0.2895, Noise=0.0069\n",
      "Iter 925: Loss=0.3229, Noise=0.0069\n",
      "Iter 926: Loss=0.2950, Noise=0.0069\n",
      "Iter 927: Loss=0.2773, Noise=0.0069\n",
      "Iter 928: Loss=0.2914, Noise=0.0069\n",
      "Iter 929: Loss=0.3101, Noise=0.0069\n",
      "Iter 930: Loss=0.3202, Noise=0.0069\n",
      "Iter 931: Loss=0.2933, Noise=0.0069\n",
      "Iter 932: Loss=0.3314, Noise=0.0069\n",
      "Iter 933: Loss=0.2874, Noise=0.0069\n",
      "Iter 934: Loss=0.3178, Noise=0.0069\n",
      "Iter 935: Loss=0.3051, Noise=0.0069\n",
      "Iter 936: Loss=0.3162, Noise=0.0069\n",
      "Iter 937: Loss=0.2740, Noise=0.0069\n",
      "Iter 938: Loss=0.3202, Noise=0.0069\n",
      "Iter 939: Loss=0.3081, Noise=0.0069\n",
      "Iter 940: Loss=0.2592, Noise=0.0069\n",
      "Iter 941: Loss=0.2610, Noise=0.0069\n",
      "Iter 942: Loss=0.2842, Noise=0.0069\n",
      "Iter 943: Loss=0.2868, Noise=0.0069\n",
      "Iter 944: Loss=0.3304, Noise=0.0069\n",
      "Iter 945: Loss=0.3001, Noise=0.0069\n",
      "Iter 946: Loss=0.2440, Noise=0.0069\n",
      "Iter 947: Loss=0.3413, Noise=0.0069\n",
      "Iter 948: Loss=0.2889, Noise=0.0069\n",
      "Iter 949: Loss=0.2836, Noise=0.0069\n",
      "Iter 950: Loss=0.3177, Noise=0.0069\n",
      "Iter 951: Loss=0.2836, Noise=0.0069\n",
      "Iter 952: Loss=0.2936, Noise=0.0069\n",
      "Iter 953: Loss=0.3100, Noise=0.0069\n",
      "Iter 954: Loss=0.2859, Noise=0.0069\n",
      "Iter 955: Loss=0.2570, Noise=0.0069\n",
      "Iter 956: Loss=0.2803, Noise=0.0069\n",
      "Iter 957: Loss=0.2955, Noise=0.0069\n",
      "Iter 958: Loss=0.2600, Noise=0.0069\n",
      "Iter 959: Loss=0.2955, Noise=0.0069\n",
      "Iter 960: Loss=0.2526, Noise=0.0069\n",
      "Iter 961: Loss=0.3059, Noise=0.0069\n",
      "Iter 962: Loss=0.2573, Noise=0.0069\n",
      "Iter 963: Loss=0.2738, Noise=0.0069\n",
      "Iter 964: Loss=0.2746, Noise=0.0069\n",
      "Iter 965: Loss=0.2648, Noise=0.0069\n",
      "Iter 966: Loss=0.2874, Noise=0.0069\n",
      "Iter 967: Loss=0.2804, Noise=0.0069\n",
      "Iter 968: Loss=0.2634, Noise=0.0070\n",
      "Iter 969: Loss=0.2624, Noise=0.0070\n",
      "Iter 970: Loss=0.2887, Noise=0.0070\n",
      "Iter 971: Loss=0.2656, Noise=0.0070\n",
      "Iter 972: Loss=0.2773, Noise=0.0070\n",
      "Iter 973: Loss=0.2772, Noise=0.0070\n",
      "Iter 974: Loss=0.2808, Noise=0.0070\n",
      "Iter 975: Loss=0.3054, Noise=0.0070\n",
      "Iter 976: Loss=0.2525, Noise=0.0070\n",
      "Iter 977: Loss=0.3110, Noise=0.0070\n",
      "Iter 978: Loss=0.3211, Noise=0.0070\n",
      "Iter 979: Loss=0.2829, Noise=0.0070\n",
      "Iter 980: Loss=0.2697, Noise=0.0070\n",
      "Iter 981: Loss=0.2932, Noise=0.0070\n",
      "Iter 982: Loss=0.2821, Noise=0.0070\n",
      "Iter 983: Loss=0.2763, Noise=0.0070\n",
      "Iter 984: Loss=0.2710, Noise=0.0070\n",
      "Iter 985: Loss=0.2781, Noise=0.0070\n",
      "Iter 986: Loss=0.2916, Noise=0.0070\n",
      "Iter 987: Loss=0.2845, Noise=0.0070\n",
      "Iter 988: Loss=0.2668, Noise=0.0070\n",
      "Iter 989: Loss=0.2552, Noise=0.0070\n",
      "Iter 990: Loss=0.2619, Noise=0.0070\n",
      "Iter 991: Loss=0.2411, Noise=0.0070\n",
      "Iter 992: Loss=0.2648, Noise=0.0070\n",
      "Iter 993: Loss=0.3159, Noise=0.0070\n",
      "Iter 994: Loss=0.2393, Noise=0.0070\n",
      "Iter 995: Loss=0.2751, Noise=0.0070\n",
      "Iter 996: Loss=0.2803, Noise=0.0070\n",
      "Iter 997: Loss=0.2618, Noise=0.0070\n",
      "Iter 998: Loss=0.2509, Noise=0.0070\n",
      "Iter 999: Loss=0.2634, Noise=0.0070\n"
     ]
    }
   ],
   "source": [
    "# Train the model\n",
    "mll = ExactMarginalLogLikelihood(model.likelihood, model)\n",
    "model.train()\n",
    "model.likelihood.train()\n",
    "optimizer = Adam(mll.parameters(), lr=0.01)\n",
    "for iteration in range(1000):\n",
    "    optimizer.zero_grad()\n",
    "    output = model(train_x)\n",
    "    loss = -mll(output, train_y)\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "    msg = \"Iter {}: Loss={:2.4f}, Noise={:2.4f}\".format(iteration, loss.item(), model.likelihood.noise.item())\n",
    "    print(msg)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSE=0.8409, Neg Log Likelihood=0.1101\n"
     ]
    }
   ],
   "source": [
    "# Test the model\n",
    "with torch.no_grad():\n",
    "    model.eval()\n",
    "    model.likelihood.eval()\n",
    "    output = model(test_x)\n",
    "    rmse = (test_y - output.mean).pow(2).mean().sqrt().item()\n",
    "    neg_log_lik = -mll(output, test_y).item()\n",
    "    print('RMSE={:2.4f}, Neg Log Likelihood={:2.4f}'.format(rmse, neg_log_lik))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:ian-keops-env]",
   "language": "python",
   "name": "conda-env-ian-keops-env-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}