meta.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<torch._C.Generator at 0x7f47a5541ed0>"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "from cls2_data.data import Dataset\n",
    "import evals2\n",
    "from util import Logger\n",
    "\n",
    "import editdistance\n",
    "import numpy as np\n",
    "import sexpdata\n",
    "import torch\n",
    "from torch import nn, optim\n",
    "from torch.optim import lr_scheduler as opt_sched\n",
    "from torch.autograd import Variable\n",
    "\n",
    "SEED = 0\n",
    "np_random = np.random.seed(SEED)\n",
    "torch.manual_seed(SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "N_EMBED = 128\n",
    "N_HIDDEN = 256\n",
    "N_BATCH = 128"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def unwrap(var):\n",
    "    return var.data.cpu().numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "class Model(nn.Module):\n",
    "    def __init__(self, dataset):\n",
    "        super().__init__()\n",
    "        self._conv_part = nn.Sequential(\n",
    "            nn.Conv2d(3, 6, 5, padding=2),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(2),\n",
    "            nn.Conv2d(6, 16, 5),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(2),\n",
    "        )\n",
    "        self._fc_part = nn.Sequential(\n",
    "            nn.Linear(16*5*5, 128),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(128, 64),\n",
    "            nn.ReLU()\n",
    "        )\n",
    "        self._pred_part = nn.Sequential(\n",
    "            nn.Linear(64, 64),\n",
    "            nn.Tanh()\n",
    "        )\n",
    "        self._loss = nn.BCEWithLogitsLoss()\n",
    "    \n",
    "    def forward(self, batch):\n",
    "        feats_in = batch.feats_in\n",
    "        feats_out = batch.feats_out\n",
    "        label_out = batch.label_out\n",
    "        if next(self.parameters()).is_cuda:\n",
    "            feats_in = feats_in.cuda()\n",
    "            feats_out = feats_out.cuda()\n",
    "            label_out = label_out.cuda()\n",
    "            \n",
    "        n_batch, n_ex, c, w, h = feats_in.shape\n",
    "        conv_in = self._conv_part(feats_in.view(n_batch * n_ex, c, w, h))\n",
    "        fc_in = self._fc_part(conv_in.view(n_batch * n_ex, 16*5*5))\n",
    "        predictor = self._pred_part(fc_in.view(n_batch, n_ex, 64).sum(dim=1))\n",
    "        \n",
    "        conv_out = self._conv_part(feats_out)\n",
    "        rep_out = self._fc_part(conv_out.view(n_batch, 16*5*5))\n",
    "        \n",
    "        score = (predictor * rep_out).sum(dim=1)\n",
    "        labels = (score > 0).float()\n",
    "        loss = self._loss(score, label_out)\n",
    "        \n",
    "        return loss, (labels == label_out).float().mean(), labels, predictor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def info(reps):\n",
    "    buckets = np.zeros((64, 30))\n",
    "    for rep in reps:\n",
    "        for i in range(len(rep)):\n",
    "            bucket = 15 + int(rep[i] * 30)\n",
    "            bucket = max(bucket, 0)\n",
    "            bucket = min(bucket, 29)\n",
    "            buckets[i, bucket] += 1\n",
    "    buckets += 1e-7\n",
    "    probs = buckets / buckets.sum(axis=1, keepdims=True)\n",
    "    logprobs = np.log(probs)\n",
    "    entropies = -(probs * logprobs).sum(axis=1)\n",
    "    return entropies.mean()\n",
    "\n",
    "EPOCH = 'epoch'\n",
    "TRN_LOSS = 'trn loss'\n",
    "TRN_ACC = 'trn acc'\n",
    "VAL_ACC = 'val acc'\n",
    "CVAL_ACC = 'cval acc'\n",
    "INFO_TX = 'I(T;X)'\n",
    "ISOM = 'isom'\n",
    "HOM = 'hom'\n",
    "CHOM = 'c_hom'\n",
    "LOG_KEYS = [EPOCH, TRN_LOSS, TRN_ACC, VAL_ACC, HOM,   ISOM,  INFO_TX]\n",
    "LOG_FMTS = ['d',   '.3f',    '.3f',   '.3f',   '.3f', '.3f', '.3f']\n",
    "\n",
    "class Composition(nn.Module):\n",
    "    def forward(self, x, y):\n",
    "        return x + y\n",
    "comp_fn = Composition()\n",
    "err_fn = evals2.CosDist()\n",
    "\n",
    "def validate(dataset, model, logger, plot_log, epoch):\n",
    "    val_batch = dataset.get_val_batch()\n",
    "    _, val_acc, _, val_reps = model(val_batch)\n",
    "    val_acc = val_acc.item()\n",
    "    logger.update(VAL_ACC, val_acc)\n",
    "    \n",
    "    cval_batch = dataset.get_cval_batch()\n",
    "    _, cval_acc, _, cval_reps = model(cval_batch)\n",
    "    cval_acc = cval_acc.item()\n",
    "    logger.update(CVAL_ACC, cval_acc)\n",
    "    \n",
    "    prim_batch = dataset.get_prim_batch()\n",
    "    _, _, _, prim_reps = model(prim_batch)\n",
    "    \n",
    "    prim_rseq = [unwrap(prim_reps[i, ...]) for i in range(prim_reps.shape[0])]\n",
    "    val_rseq = [unwrap(val_reps[i, ...]) for i in range(val_reps.shape[0])]\n",
    "    cval_rseq = [unwrap(cval_reps[i, ...]) for i in range(cval_reps.shape[0])]\n",
    "    \n",
    "    comp = evals2.evaluate(\n",
    "        prim_rseq + val_rseq, prim_batch.lf + val_batch.lf,\n",
    "        comp_fn, err_fn, quiet=True)\n",
    "    logger.update(HOM, np.mean(comp))\n",
    "    \n",
    "    #ccomp = evals2.evaluate(\n",
    "    #    prim_rseq + cval_rseq, prim_batch.lf + cval_batch.lf,\n",
    "    #    comp_fn, err_fn)[-len(cval_rseq):]\n",
    "    #logger.update(CHOM, np.mean(ccomp))\n",
    "    \n",
    "    #logger.update(ISOM, eval_isom_tree(unwrap(val_reps), val_batch.lf))\n",
    "    #info_tx = info(unwrap(nn.functional.tanh(val_reps)))\n",
    "    info_tx = info(unwrap(val_reps))\n",
    "    logger.update(INFO_TX, info_tx)\n",
    "    \n",
    "    plot_log.append((epoch, info_tx, np.mean(comp), val_acc))\n",
    "    \n",
    "    return val_acc\n",
    "\n",
    "def train(dataset, model):\n",
    "    #model.cuda()\n",
    "    \n",
    "    opt = optim.Adam(model.parameters(), lr=1e-3)\n",
    "    sched = opt_sched.ReduceLROnPlateau(opt, factor=0.5, verbose=True, mode='max')\n",
    "    logger = Logger(LOG_KEYS, LOG_FMTS, width=10)\n",
    "    logger.begin()\n",
    "    \n",
    "    val_acc = validate(dataset, model, logger, [], -1)\n",
    "    logger.print()\n",
    "    \n",
    "    plot_log = []\n",
    "    for i in range(20):\n",
    "        trn_loss = 0\n",
    "        trn_acc = 0\n",
    "        for j in range(100):\n",
    "            batch = dataset.get_train_batch(N_BATCH)\n",
    "            loss, acc, _, _ = model(batch)\n",
    "            opt.zero_grad()\n",
    "            loss.backward()\n",
    "            opt.step()\n",
    "            trn_loss += loss.item()\n",
    "            trn_acc += acc.item()\n",
    "            \n",
    "            #if i == 0 and j % 10 == 0:\n",
    "            #    validate(dataset, model, logger, [], -1)\n",
    "            #    logger.print()\n",
    "                \n",
    "        trn_loss /= 100\n",
    "        trn_acc /= 100\n",
    "        \n",
    "        logger.update(EPOCH, i)\n",
    "        logger.update(TRN_LOSS, trn_loss)\n",
    "        logger.update(TRN_ACC, trn_acc)\n",
    "        val_acc = validate(dataset, model, logger, plot_log, i)\n",
    "        sched.step(val_acc)\n",
    "        logger.print()\n",
    "    return plot_log"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n",
      "|            |            |            |      0.496 |      0.015 |            |      0.757 |\n",
      "|          0 |      0.663 |      0.600 |      0.668 |      0.108 |            |      1.955 |\n",
      "|          1 |      0.630 |      0.655 |      0.688 |      0.150 |            |      2.453 |\n",
      "|          2 |      0.613 |      0.678 |      0.704 |      0.159 |            |      2.743 |\n",
      "|          3 |      0.598 |      0.683 |      0.720 |      0.181 |            |      2.802 |\n",
      "|          4 |      0.583 |      0.694 |      0.706 |      0.181 |            |      2.848 |\n",
      "|          5 |      0.581 |      0.695 |      0.700 |      0.197 |            |      2.866 |\n",
      "|          6 |      0.569 |      0.709 |      0.742 |      0.197 |            |      2.869 |\n",
      "|          7 |      0.555 |      0.720 |      0.748 |      0.198 |            |      2.898 |\n",
      "|          8 |      0.548 |      0.723 |      0.724 |      0.182 |            |      2.850 |\n",
      "|          9 |      0.537 |      0.730 |      0.738 |      0.192 |            |      2.882 |\n",
      "|         10 |      0.526 |      0.736 |      0.744 |      0.186 |            |      2.828 |\n",
      "|         11 |      0.504 |      0.755 |      0.734 |      0.184 |            |      2.833 |\n",
      "|         12 |      0.492 |      0.758 |      0.758 |      0.175 |            |      2.778 |\n",
      "|         13 |      0.481 |      0.771 |      0.740 |      0.180 |            |      2.640 |\n",
      "|         14 |      0.467 |      0.778 |      0.744 |      0.170 |            |      2.696 |\n",
      "|         15 |      0.443 |      0.791 |      0.728 |      0.174 |            |      2.609 |\n",
      "|         16 |      0.429 |      0.804 |      0.742 |      0.175 |            |      2.582 |\n",
      "|         17 |      0.429 |      0.803 |      0.716 |      0.194 |            |      2.625 |\n",
      "|         18 |      0.397 |      0.822 |      0.720 |      0.184 |            |      2.468 |\n",
      "|         19 |      0.373 |      0.834 |      0.736 |      0.188 |            |      2.377 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n",
      "|            |            |            |      0.474 |      0.011 |            |      1.002 |\n",
      "|          0 |      0.672 |      0.569 |      0.612 |      0.084 |            |      1.984 |\n",
      "|          1 |      0.641 |      0.626 |      0.710 |      0.134 |            |      2.389 |\n",
      "|          2 |      0.614 |      0.672 |      0.700 |      0.136 |            |      2.370 |\n",
      "|          3 |      0.611 |      0.676 |      0.690 |      0.133 |            |      2.291 |\n",
      "|          4 |      0.612 |      0.670 |      0.704 |      0.139 |            |      2.499 |\n",
      "|          5 |      0.594 |      0.684 |      0.730 |      0.115 |            |      2.671 |\n",
      "|          6 |      0.575 |      0.700 |      0.726 |      0.135 |            |      2.716 |\n",
      "|          7 |      0.560 |      0.711 |      0.748 |      0.134 |            |      2.644 |\n",
      "|          8 |      0.538 |      0.732 |      0.714 |      0.123 |            |      2.610 |\n",
      "|          9 |      0.529 |      0.736 |      0.720 |      0.114 |            |      2.585 |\n",
      "|         10 |      0.515 |      0.747 |      0.742 |      0.111 |            |      2.475 |\n",
      "|         11 |      0.498 |      0.754 |      0.750 |      0.118 |            |      2.533 |\n",
      "|         12 |      0.485 |      0.764 |      0.748 |      0.109 |            |      2.439 |\n",
      "|         13 |      0.469 |      0.775 |      0.730 |      0.108 |            |      2.347 |\n",
      "|         14 |      0.450 |      0.791 |      0.736 |      0.112 |            |      2.373 |\n",
      "|         15 |      0.427 |      0.807 |      0.738 |      0.116 |            |      2.363 |\n",
      "|         16 |      0.404 |      0.815 |      0.720 |      0.117 |            |      2.323 |\n",
      "|         17 |      0.385 |      0.828 |      0.748 |      0.119 |            |      2.201 |\n",
      "|         18 |      0.369 |      0.839 |      0.736 |      0.122 |            |      2.166 |\n",
      "|         19 |      0.346 |      0.852 |      0.720 |      0.125 |            |      2.174 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n",
      "|            |            |            |      0.496 |      0.012 |            |      0.716 |\n",
      "|          0 |      0.653 |      0.608 |      0.688 |      0.127 |            |      2.202 |\n",
      "|          1 |      0.618 |      0.665 |      0.704 |      0.128 |            |      2.653 |\n",
      "|          2 |      0.594 |      0.689 |      0.682 |      0.144 |            |      2.738 |\n",
      "|          3 |      0.588 |      0.691 |      0.708 |      0.140 |            |      2.771 |\n",
      "|          4 |      0.577 |      0.700 |      0.728 |      0.155 |            |      2.852 |\n",
      "|          5 |      0.553 |      0.718 |      0.744 |      0.157 |            |      2.799 |\n",
      "|          6 |      0.552 |      0.719 |      0.728 |      0.147 |            |      2.841 |\n",
      "|          7 |      0.536 |      0.733 |      0.716 |      0.142 |            |      2.703 |\n",
      "|          8 |      0.525 |      0.739 |      0.708 |      0.143 |            |      2.829 |\n",
      "|          9 |      0.503 |      0.758 |      0.752 |      0.150 |            |      2.797 |\n",
      "|         10 |      0.483 |      0.768 |      0.760 |      0.150 |            |      2.714 |\n",
      "|         11 |      0.462 |      0.785 |      0.766 |      0.150 |            |      2.549 |\n",
      "|         12 |      0.454 |      0.783 |      0.776 |      0.144 |            |      2.513 |\n",
      "|         13 |      0.436 |      0.795 |      0.766 |      0.137 |            |      2.433 |\n",
      "|         14 |      0.414 |      0.815 |      0.770 |      0.139 |            |      2.459 |\n",
      "|         15 |      0.392 |      0.826 |      0.756 |      0.141 |            |      2.288 |\n",
      "|         16 |      0.369 |      0.835 |      0.774 |      0.140 |            |      2.220 |\n",
      "|         17 |      0.351 |      0.847 |      0.760 |      0.149 |            |      2.172 |\n",
      "|         18 |      0.337 |      0.853 |      0.762 |      0.146 |            |      2.129 |\n",
      "|         19 |      0.313 |      0.870 |      0.744 |      0.156 |            |      2.124 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n",
      "|            |            |            |      0.498 |      0.014 |            |      0.969 |\n",
      "|          0 |      0.652 |      0.621 |      0.694 |      0.135 |            |      2.261 |\n",
      "|          1 |      0.614 |      0.682 |      0.698 |      0.168 |            |      2.801 |\n",
      "|          2 |      0.594 |      0.688 |      0.704 |      0.157 |            |      2.732 |\n",
      "|          3 |      0.578 |      0.703 |      0.714 |      0.149 |            |      2.765 |\n",
      "|          4 |      0.574 |      0.701 |      0.724 |      0.157 |            |      2.824 |\n",
      "|          5 |      0.564 |      0.711 |      0.720 |      0.163 |            |      2.847 |\n",
      "|          6 |      0.561 |      0.715 |      0.716 |      0.152 |            |      2.818 |\n",
      "|          7 |      0.541 |      0.728 |      0.688 |      0.162 |            |      2.852 |\n",
      "|          8 |      0.530 |      0.733 |      0.730 |      0.132 |            |      2.681 |\n",
      "|          9 |      0.513 |      0.750 |      0.758 |      0.155 |            |      2.819 |\n",
      "|         10 |      0.485 |      0.768 |      0.742 |      0.157 |            |      2.701 |\n",
      "|         11 |      0.477 |      0.772 |      0.726 |      0.159 |            |      2.638 |\n",
      "|         12 |      0.454 |      0.790 |      0.746 |      0.157 |            |      2.560 |\n",
      "|         13 |      0.432 |      0.801 |      0.726 |      0.171 |            |      2.566 |\n",
      "|         14 |      0.416 |      0.810 |      0.752 |      0.172 |            |      2.492 |\n",
      "|         15 |      0.391 |      0.826 |      0.718 |      0.169 |            |      2.516 |\n",
      "|         16 |      0.369 |      0.836 |      0.752 |      0.178 |            |      2.307 |\n",
      "|         17 |      0.345 |      0.850 |      0.736 |      0.177 |            |      2.218 |\n",
      "|         18 |      0.320 |      0.864 |      0.738 |      0.187 |            |      2.268 |\n",
      "|         19 |      0.308 |      0.870 |      0.734 |      0.184 |            |      2.163 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "|            |            |            |      0.504 |      0.010 |            |      0.756 |\n",
      "|          0 |      0.673 |      0.546 |      0.594 |      0.116 |            |      1.507 |\n",
      "|          1 |      0.662 |      0.580 |      0.592 |      0.096 |            |      1.319 |\n",
      "|          2 |      0.660 |      0.590 |      0.596 |      0.100 |            |      1.923 |\n",
      "|          3 |      0.650 |      0.606 |      0.642 |      0.110 |            |      2.512 |\n",
      "|          4 |      0.629 |      0.642 |      0.672 |      0.123 |            |      2.712 |\n",
      "|          5 |      0.602 |      0.672 |      0.700 |      0.165 |            |      2.851 |\n",
      "|          6 |      0.579 |      0.704 |      0.724 |      0.166 |            |      2.951 |\n",
      "|          7 |      0.555 |      0.718 |      0.738 |      0.159 |            |      2.906 |\n",
      "|          8 |      0.546 |      0.722 |      0.744 |      0.143 |            |      2.880 |\n",
      "|          9 |      0.530 |      0.738 |      0.742 |      0.139 |            |      2.832 |\n",
      "|         10 |      0.525 |      0.738 |      0.714 |      0.145 |            |      2.825 |\n",
      "|         11 |      0.493 |      0.760 |      0.734 |      0.162 |            |      2.775 |\n",
      "|         12 |      0.490 |      0.766 |      0.728 |      0.147 |            |      2.757 |\n",
      "|         13 |      0.479 |      0.771 |      0.726 |      0.144 |            |      2.672 |\n",
      "|         14 |      0.457 |      0.784 |      0.708 |      0.152 |            |      2.662 |\n",
      "|         15 |      0.438 |      0.795 |      0.734 |      0.149 |            |      2.513 |\n",
      "|         16 |      0.418 |      0.809 |      0.714 |      0.147 |            |      2.503 |\n",
      "|         17 |      0.397 |      0.816 |      0.714 |      0.147 |            |      2.262 |\n",
      "|         18 |      0.382 |      0.827 |      0.710 |      0.153 |            |      2.290 |\n",
      "Epoch    19: reducing learning rate of group 0 to 5.0000e-04.\n",
      "|         19 |      0.360 |      0.845 |      0.728 |      0.158 |            |      2.276 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n",
      "|            |            |            |      0.496 |      0.009 |            |      0.693 |\n",
      "|          0 |      0.667 |      0.578 |      0.638 |      0.132 |            |      2.495 |\n",
      "|          1 |      0.637 |      0.629 |      0.662 |      0.154 |            |      2.496 |\n",
      "|          2 |      0.622 |      0.645 |      0.648 |      0.146 |            |      2.561 |\n",
      "|          3 |      0.616 |      0.659 |      0.686 |      0.160 |            |      2.688 |\n",
      "|          4 |      0.591 |      0.687 |      0.706 |      0.164 |            |      2.849 |\n",
      "|          5 |      0.555 |      0.718 |      0.724 |      0.160 |            |      2.842 |\n",
      "|          6 |      0.542 |      0.729 |      0.708 |      0.146 |            |      2.801 |\n",
      "|          7 |      0.545 |      0.726 |      0.734 |      0.138 |            |      2.752 |\n",
      "|          8 |      0.537 |      0.731 |      0.714 |      0.137 |            |      2.733 |\n",
      "|          9 |      0.530 |      0.734 |      0.708 |      0.139 |            |      2.690 |\n",
      "|         10 |      0.521 |      0.739 |      0.720 |      0.141 |            |      2.687 |\n",
      "|         11 |      0.512 |      0.749 |      0.718 |      0.144 |            |      2.675 |\n",
      "|         12 |      0.502 |      0.757 |      0.744 |      0.142 |            |      2.613 |\n",
      "|         13 |      0.483 |      0.769 |      0.730 |      0.136 |            |      2.589 |\n",
      "|         14 |      0.458 |      0.787 |      0.732 |      0.143 |            |      2.437 |\n",
      "|         15 |      0.451 |      0.794 |      0.730 |      0.136 |            |      2.495 |\n",
      "|         16 |      0.428 |      0.801 |      0.724 |      0.156 |            |      2.471 |\n",
      "|         17 |      0.414 |      0.812 |      0.718 |      0.150 |            |      2.389 |\n",
      "|         18 |      0.381 |      0.829 |      0.724 |      0.148 |            |      2.307 |\n",
      "|         19 |      0.362 |      0.842 |      0.726 |      0.152 |            |      2.234 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n",
      "|            |            |            |      0.504 |      0.013 |            |      0.930 |\n",
      "|          0 |      0.661 |      0.602 |      0.670 |      0.114 |            |      2.007 |\n",
      "|          1 |      0.651 |      0.624 |      0.696 |      0.161 |            |      2.252 |\n",
      "|          2 |      0.633 |      0.651 |      0.684 |      0.188 |            |      2.667 |\n",
      "|          3 |      0.599 |      0.686 |      0.692 |      0.165 |            |      2.813 |\n",
      "|          4 |      0.576 |      0.708 |      0.700 |      0.174 |            |      2.869 |\n",
      "|          5 |      0.568 |      0.707 |      0.716 |      0.162 |            |      2.866 |\n",
      "|          6 |      0.558 |      0.718 |      0.714 |      0.175 |            |      2.857 |\n",
      "|          7 |      0.545 |      0.722 |      0.728 |      0.169 |            |      2.858 |\n",
      "|          8 |      0.535 |      0.733 |      0.706 |      0.151 |            |      2.784 |\n",
      "|          9 |      0.512 |      0.752 |      0.720 |      0.143 |            |      2.642 |\n",
      "|         10 |      0.484 |      0.762 |      0.722 |      0.140 |            |      2.630 |\n",
      "|         11 |      0.472 |      0.775 |      0.728 |      0.140 |            |      2.494 |\n",
      "|         12 |      0.446 |      0.790 |      0.720 |      0.138 |            |      2.396 |\n",
      "|         13 |      0.441 |      0.795 |      0.710 |      0.132 |            |      2.375 |\n",
      "|         14 |      0.409 |      0.816 |      0.748 |      0.132 |            |      2.269 |\n",
      "|         15 |      0.391 |      0.827 |      0.728 |      0.133 |            |      2.196 |\n",
      "|         16 |      0.374 |      0.836 |      0.724 |      0.140 |            |      2.234 |\n",
      "|         17 |      0.345 |      0.849 |      0.710 |      0.145 |            |      2.142 |\n",
      "|         18 |      0.331 |      0.858 |      0.712 |      0.138 |            |      2.152 |\n",
      "|         19 |      0.301 |      0.874 |      0.724 |      0.154 |            |      1.956 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n",
      "|            |            |            |      0.484 |      0.015 |            |      0.742 |\n",
      "|          0 |      0.647 |      0.624 |      0.696 |      0.122 |            |      2.146 |\n",
      "|          1 |      0.626 |      0.660 |      0.674 |      0.163 |            |      2.643 |\n",
      "|          2 |      0.595 |      0.688 |      0.692 |      0.181 |            |      2.760 |\n",
      "|          3 |      0.576 |      0.702 |      0.684 |      0.182 |            |      2.801 |\n",
      "|          4 |      0.579 |      0.701 |      0.706 |      0.167 |            |      2.788 |\n",
      "|          5 |      0.575 |      0.704 |      0.698 |      0.167 |            |      2.813 |\n",
      "|          6 |      0.564 |      0.713 |      0.710 |      0.160 |            |      2.829 |\n",
      "|          7 |      0.555 |      0.724 |      0.702 |      0.147 |            |      2.797 |\n",
      "|          8 |      0.534 |      0.733 |      0.724 |      0.160 |            |      2.867 |\n",
      "|          9 |      0.521 |      0.743 |      0.738 |      0.157 |            |      2.825 |\n",
      "|         10 |      0.510 |      0.749 |      0.742 |      0.143 |            |      2.604 |\n",
      "|         11 |      0.491 |      0.761 |      0.716 |      0.136 |            |      2.597 |\n",
      "|         12 |      0.475 |      0.773 |      0.716 |      0.147 |            |      2.521 |\n",
      "|         13 |      0.456 |      0.789 |      0.728 |      0.148 |            |      2.458 |\n",
      "|         14 |      0.439 |      0.799 |      0.702 |      0.144 |            |      2.433 |\n",
      "|         15 |      0.421 |      0.807 |      0.702 |      0.157 |            |      2.342 |\n",
      "|         16 |      0.400 |      0.816 |      0.728 |      0.152 |            |      2.303 |\n",
      "|         17 |      0.384 |      0.828 |      0.722 |      0.153 |            |      2.197 |\n",
      "|         18 |      0.372 |      0.837 |      0.698 |      0.172 |            |      2.286 |\n",
      "|         19 |      0.329 |      0.860 |      0.722 |      0.174 |            |      2.222 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "|            |            |            |      0.502 |      0.023 |            |      1.024 |\n",
      "|          0 |      0.656 |      0.597 |      0.688 |      0.141 |            |      2.274 |\n",
      "|          1 |      0.631 |      0.656 |      0.698 |      0.138 |            |      2.462 |\n",
      "|          2 |      0.602 |      0.685 |      0.718 |      0.148 |            |      2.755 |\n",
      "|          3 |      0.583 |      0.697 |      0.718 |      0.156 |            |      2.752 |\n",
      "|          4 |      0.580 |      0.703 |      0.720 |      0.158 |            |      2.787 |\n",
      "|          5 |      0.574 |      0.704 |      0.720 |      0.158 |            |      2.776 |\n",
      "|          6 |      0.563 |      0.710 |      0.728 |      0.154 |            |      2.795 |\n",
      "|          7 |      0.557 |      0.718 |      0.742 |      0.167 |            |      2.811 |\n",
      "|          8 |      0.538 |      0.733 |      0.738 |      0.143 |            |      2.800 |\n",
      "|          9 |      0.522 |      0.744 |      0.732 |      0.144 |            |      2.807 |\n",
      "|         10 |      0.509 |      0.748 |      0.734 |      0.150 |            |      2.806 |\n",
      "|         11 |      0.497 |      0.757 |      0.720 |      0.163 |            |      2.823 |\n",
      "|         12 |      0.476 |      0.776 |      0.730 |      0.163 |            |      2.585 |\n",
      "|         13 |      0.460 |      0.783 |      0.706 |      0.168 |            |      2.682 |\n",
      "|         14 |      0.447 |      0.793 |      0.698 |      0.168 |            |      2.614 |\n",
      "|         15 |      0.412 |      0.811 |      0.728 |      0.177 |            |      2.568 |\n",
      "|         16 |      0.389 |      0.822 |      0.720 |      0.168 |            |      2.463 |\n",
      "|         17 |      0.374 |      0.831 |      0.708 |      0.179 |            |      2.406 |\n",
      "Epoch    18: reducing learning rate of group 0 to 5.0000e-04.\n",
      "|         18 |      0.342 |      0.854 |      0.728 |      0.185 |            |      2.415 |\n",
      "|         19 |      0.312 |      0.868 |      0.718 |      0.185 |            |      2.224 |\n",
      "|      epoch |   trn loss |    trn acc |    val acc |        hom |       isom |     I(T;X) |\n",
      "|            |            |            |      0.502 |      0.012 |            |      0.654 |\n",
      "|          0 |      0.672 |      0.573 |      0.644 |      0.120 |            |      2.091 |\n",
      "|          1 |      0.643 |      0.639 |      0.664 |      0.153 |            |      2.377 |\n",
      "|          2 |      0.623 |      0.665 |      0.676 |      0.139 |            |      2.419 |\n",
      "|          3 |      0.619 |      0.668 |      0.700 |      0.128 |            |      2.517 |\n",
      "|          4 |      0.600 |      0.681 |      0.704 |      0.148 |            |      2.764 |\n",
      "|          5 |      0.576 |      0.699 |      0.688 |      0.149 |            |      2.752 |\n",
      "|          6 |      0.568 |      0.710 |      0.714 |      0.153 |            |      2.801 |\n",
      "|          7 |      0.556 |      0.717 |      0.752 |      0.147 |            |      2.775 |\n",
      "|          8 |      0.526 |      0.742 |      0.738 |      0.152 |            |      2.791 |\n",
      "|          9 |      0.522 |      0.741 |      0.710 |      0.158 |            |      2.743 |\n",
      "|         10 |      0.484 |      0.770 |      0.732 |      0.145 |            |      2.682 |\n",
      "|         11 |      0.473 |      0.777 |      0.732 |      0.165 |            |      2.732 |\n",
      "|         12 |      0.442 |      0.795 |      0.760 |      0.153 |            |      2.644 |\n",
      "|         13 |      0.420 |      0.810 |      0.740 |      0.156 |            |      2.479 |\n",
      "|         14 |      0.399 |      0.820 |      0.744 |      0.159 |            |      2.299 |\n",
      "|         15 |      0.377 |      0.829 |      0.756 |      0.151 |            |      2.324 |\n",
      "|         16 |      0.348 |      0.849 |      0.746 |      0.158 |            |      2.235 |\n",
      "|         17 |      0.319 |      0.867 |      0.744 |      0.155 |            |      2.211 |\n",
      "|         18 |      0.294 |      0.877 |      0.734 |      0.161 |            |      2.071 |\n",
      "|         19 |      0.294 |      0.881 |      0.754 |      0.163 |            |      2.065 |\n"
     ]
    }
   ],
   "source": [
    "logs = []\n",
    "dataset = Dataset()\n",
    "for i in range(10):\n",
    "    model = Model(dataset)\n",
    "    log = train(dataset, model)\n",
    "    logs.append(log)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(0.707063088342884, 1.7203824835743277e-17)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4728b2cd30>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAVkAAAFPCAYAAADnfb1dAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nqAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4722fd5f28>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "from pandas import DataFrame\n",
    "import seaborn as sns\n",
    "import scipy\n",
    "\n",
    "sns.set(font_scale=1.5)\n",
    "sns.set_style(\"ticks\", {'font.family': 'serif'})\n",
    "plt.tight_layout()\n",
    "#cmap = sns.color_palette(\"coolwarm\", 10)\n",
    "\n",
    "best_stops = [np.argmax(l, axis=0)[3] for l in logs]\n",
    "\n",
    "my_logs = logs\n",
    "my_logs = [my_logs[i][:best_stops[i]] for i in range(len(my_logs))]\n",
    "log = sum(my_logs, [])\n",
    "data = DataFrame(np.asarray(log), columns=['epoch', 'I(θ;X)', 'TRE', 'val'])\n",
    "sns.lmplot(x='I(θ;X)', y='TRE', data=data)\n",
    "print(scipy.stats.pearsonr(data['I(θ;X)'], data['TRE']))\n",
    "plt.savefig('meta2_scatter_all.pdf', format='pdf')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(0.7138623817683064, 1.2272481355539673e-08)\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAU8AAAFPCAYAAADNzUzyAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nxAEUnoQQ4gAKT0IIccD/AwZ9nZb96kdbAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4722ea0860>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#my_logs = [l[10:] for l in my_logs]\n",
    "#my_logs = [l[10:-2] for l in logs]\n",
    "info_maxes = [np.argmax(l, axis=0)[1] for l in logs]\n",
    "my_logs = logs\n",
    "my_logs = [my_logs[i][info_maxes[i]:best_stops[i]] for i in range(len(my_logs))]\n",
    "#my_logs = [my_logs[i][0:info_maxes[i]] for i in range(len(my_logs))]\n",
    "log = sum(my_logs, [])\n",
    "data = DataFrame(np.asarray(log), columns=['epoch', 'I(θ;X)', 'TRE', 'val'])\n",
    "sns.lmplot(x='I(θ;X)', y='TRE', data=data)\n",
    "print(scipy.stats.pearsonr(data['I(θ;X)'], data['TRE']))\n",
    "plt.savefig('meta2_scatter_last.pdf', format='pdf')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(0.7218689328490042, 0.003556924610567421)\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUkAAAFPCAYAAADA0zy1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f4722bca828>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "use_log = 6\n",
    "log = logs[use_log][:best_stops[use_log]]\n",
    "data = DataFrame(np.asarray(log), columns=['epoch', 'I(θ;X)', 'TRE', 'val'])\n",
    "g = sns.lmplot(x='I(θ;X)', y='TRE', data=data, fit_reg=False, scatter=False)\n",
    "plt.plot(data['I(θ;X)'], data['TRE'], color='lightgray', zorder=-1)\n",
    "plt.scatter(data['I(θ;X)'], data['TRE'], c=data['epoch'].astype(int), cmap='viridis')\n",
    "print(scipy.stats.pearsonr(data['I(θ;X)'], data['TRE']))\n",
    "plt.colorbar(ticks=range(0, 20, 2))\n",
    "plt.savefig('meta2_scatter_1.pdf', format='pdf')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}