proto_true.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import math\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "import torch.autograd as autograd\n",
    "import torchvision.transforms as transforms\n",
    "from torch.autograd import Variable\n",
    "from torch.nn.parameter import Parameter\n",
    "%matplotlib inline\n",
    "import pylab as pl\n",
    "from IPython import display\n",
    "import time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "class Block(nn.Module):\n",
    "    def __init__(self, insize, outsize):\n",
    "        super(Block, self).__init__()\n",
    "        self.layers = nn.Sequential(\n",
    "            nn.Conv2d(insize, outsize, kernel_size=3, padding=1),\n",
    "            nn.BatchNorm2d(outsize),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(2)\n",
    "        )\n",
    "        \n",
    "    def forward(self, inp):\n",
    "        return self.layers(inp)\n",
    "\n",
    "class ENCODER(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(ENCODER, self).__init__()\n",
    "        self.block1 = Block(3,64)\n",
    "        self.block2 = Block(64,64)\n",
    "        self.block3 = Block(64,64)\n",
    "        self.block4 = Block(64,64)\n",
    "#         self.final = nn.Conv2d(64,64, kernel_size=1, padding=0)\n",
    "        \n",
    "    def forward(self, inp):\n",
    "        out = self.block1(inp)\n",
    "        out = self.block2(out)\n",
    "        out = self.block3(out)\n",
    "        out = self.block4(out)\n",
    "#         out = self.final(out)\n",
    "        return out.view(out.size(0),-1)\n",
    "    \n",
    "class PREDICTOR(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(PREDICTOR, self).__init__()\n",
    "        self.sm = nn.LogSoftmax()\n",
    "        \n",
    "    def forward(self, support, inp, way):\n",
    "        assert support.size(0)%way == 0,\"Error: classes are not the same size!\"\n",
    "        support = support.view(way,support.size(0)//way,-1)\n",
    "        centroids = torch.mean(support, 1).view(support.size(0),-1)\n",
    "#         temp1 = torch.stack([centroids]*inp.size(0))\n",
    "#         temp2 = torch.stack([inp]*centroids.size(0)).transpose(0,1)\n",
    "#         print(temp1.size(),temp2.size())\n",
    "        distmat = torch.sum((centroids.unsqueeze(0)-inp.unsqueeze(1))**2,2).squeeze()*-1\n",
    "        probs = self.sm(distmat)\n",
    "        return probs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "trainset = torch.load('mIN_train.pth')\n",
    "testset = torch.load('mIN_test.pth')\n",
    "valset = torch.load('mIN_val.pth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "standardize = transforms.Compose([\n",
    "            transforms.ToTensor(),\n",
    "            transforms.Normalize(mean=[.485,.456,.406],std=[.229,.224,.225])\n",
    "            ])\n",
    "alter = transforms.Compose([\n",
    "            transforms.ToPILImage(),\n",
    "            transforms.RandomHorizontalFlip(),\n",
    "            transforms.RandomCrop(84,padding=10),\n",
    "            standardize\n",
    "            ])\n",
    "\n",
    "def batchmaker(way,trainshot,testshot,theset,alterful=False):\n",
    "    classes = np.random.choice(len(theset),way)\n",
    "    if alterful:\n",
    "        li = [torch.cat([alter(theset[cl][i]).view(1,3,84,84) for i in \n",
    "                         np.random.choice(600,trainshot+testshot)],dim=0).float()\n",
    "              for cl in classes]\n",
    "    else:\n",
    "        li = [torch.cat([standardize(theset[cl][i]).view(1,3,84,84) for i in \n",
    "                         np.random.choice(600,trainshot+testshot)],dim=0).float()\n",
    "              for cl in classes]\n",
    "    support = torch.cat([t[:trainshot,:,:,:] for t in li],dim=0)\n",
    "    stargs = torch.LongTensor([i//trainshot for i in range(trainshot*way)])\n",
    "    query = torch.cat([t[trainshot:,:,:,:] for t in li],dim=0)\n",
    "    qtargs = torch.LongTensor([i//testshot for i in range(testshot*way)])\n",
    "#     print(support.size(),query.size())\n",
    "    return(Variable(support).cuda(),\n",
    "           Variable(query, volatile=(not alterful)).cuda(),\n",
    "           Variable(qtargs, volatile=(not alterful)).cuda(),\n",
    "           Variable(stargs).cuda()\n",
    "          )\n",
    "\n",
    "def evaluate(embed,model,criterion,way,trainshot,testshot,reps,theset):\n",
    "    losstracker = 0\n",
    "    acctracker = 0\n",
    "    acclist = []\n",
    "    for i in range(reps):\n",
    "        s,q,qt,st = batchmaker(way,trainshot,testshot,theset)\n",
    "        embeds = embed(s)\n",
    "        qembeds = embed(q)\n",
    "        preds = model(embeds,qembeds,way)\n",
    "        \n",
    "        # Val loss\n",
    "        loss = criterion(preds,qt)\n",
    "        losstracker += loss.data[0]\n",
    "        \n",
    "        # Val accuracy\n",
    "        _,bins = torch.max(preds,1)\n",
    "        acc = torch.sum(torch.eq(bins,qt)).data[0]/testshot/way\n",
    "        acctracker += acc\n",
    "        acclist.append(acc)\n",
    "    return losstracker/reps,acctracker/reps,np.var(acclist)\n",
    "\n",
    "way = 20\n",
    "evalway = 5\n",
    "trainshot = 5\n",
    "testshot = 15\n",
    "reps = 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "torch.cuda.set_device(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "113088  parameters in neural net.\n"
     ]
    }
   ],
   "source": [
    "embed = ENCODER().cuda()\n",
    "model = PREDICTOR().cuda()\n",
    "losstracker,evallosstracker,evalacctracker = [],[],[]\n",
    "optimizer = None\n",
    "nweights = sum([i.numel() for i in list(embed.parameters())])\n",
    "print(nweights,\" parameters in neural net.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA20AAAE/CAYAAADVKysfAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nMEObJEmSJFWYoU2SJEmSKuz/ATZfaf9akuXbAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f6ee20f99b0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train loss is: 1.801734420657158\n",
      "Validation accuracy is: 0.6213333333333332\n",
      "Validation loss is: 0.9442143198847771\n",
      "\n",
      "666.3798024654388\n",
      "3200\n",
      "3210\n",
      "3220\n",
      "3230\n",
      "3240\n",
      "3250\n",
      "3260\n",
      "3270\n",
      "3280\n",
      "3290\n",
      "3300\n",
      "3310\n",
      "3320\n",
      "3330\n",
      "3340\n",
      "3350\n",
      "3360\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-10-5f974ca72d16>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     11\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     12\u001b[0m     \u001b[0;31m# Build batch\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 13\u001b[0;31m     \u001b[0msupport\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mquery\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtargs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0m_\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mbatchmaker\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mway\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mtrainshot\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mtestshot\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mtrainset\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0malterful\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     14\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     15\u001b[0m     \u001b[0;31m# Predict\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m<ipython-input-4-d4a869db6b5e>\u001b[0m in \u001b[0;36mbatchmaker\u001b[0;34m(way, trainshot, testshot, theset, alterful)\u001b[0m\n\u001b[1;32m     15\u001b[0m         li = [torch.cat([alter(theset[cl][i]).view(1,3,84,84) for i in \n\u001b[1;32m     16\u001b[0m                          np.random.choice(600,trainshot+testshot)],dim=0).float()\n\u001b[0;32m---> 17\u001b[0;31m               for cl in classes]\n\u001b[0m\u001b[1;32m     18\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     19\u001b[0m         li = [torch.cat([standardize(theset[cl][i]).view(1,3,84,84) for i in \n",
      "\u001b[0;32m<ipython-input-4-d4a869db6b5e>\u001b[0m in \u001b[0;36m<listcomp>\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m     15\u001b[0m         li = [torch.cat([alter(theset[cl][i]).view(1,3,84,84) for i in \n\u001b[1;32m     16\u001b[0m                          np.random.choice(600,trainshot+testshot)],dim=0).float()\n\u001b[0;32m---> 17\u001b[0;31m               for cl in classes]\n\u001b[0m\u001b[1;32m     18\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     19\u001b[0m         li = [torch.cat([standardize(theset[cl][i]).view(1,3,84,84) for i in \n",
      "\u001b[0;32m<ipython-input-4-d4a869db6b5e>\u001b[0m in \u001b[0;36m<listcomp>\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m     13\u001b[0m     \u001b[0mclasses\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrandom\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mchoice\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtheset\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mway\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0malterful\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 15\u001b[0;31m         li = [torch.cat([alter(theset[cl][i]).view(1,3,84,84) for i in \n\u001b[0m\u001b[1;32m     16\u001b[0m                          np.random.choice(600,trainshot+testshot)],dim=0).float()\n\u001b[1;32m     17\u001b[0m               for cl in classes]\n",
      "\u001b[0;32m~/anaconda3/lib/python3.6/site-packages/torchvision/transforms.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, img)\u001b[0m\n\u001b[1;32m     27\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m__call__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mimg\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     28\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mt\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtransforms\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 29\u001b[0;31m             \u001b[0mimg\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mimg\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     30\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mimg\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     31\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/lib/python3.6/site-packages/torchvision/transforms.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, img)\u001b[0m\n\u001b[1;32m     27\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m__call__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mimg\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     28\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mt\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtransforms\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 29\u001b[0;31m             \u001b[0mimg\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mimg\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     30\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mimg\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     31\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/lib/python3.6/site-packages/torchvision/transforms.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, pic)\u001b[0m\n\u001b[1;32m     48\u001b[0m             \u001b[0mimg\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfrom_numpy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0marray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpic\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mint16\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcopy\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     49\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 50\u001b[0;31m             \u001b[0mimg\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mByteTensor\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mByteStorage\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfrom_buffer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpic\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtobytes\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     51\u001b[0m         \u001b[0;31m# PIL image mode: 1, L, P, I, F, RGB, YCbCr, RGBA, CMYK\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     52\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mpic\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmode\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;34m'YCbCr'\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "# TRAIN!\n",
    "criterion = nn.NLLLoss().cuda()\n",
    "vbity = 200\n",
    "epoch = 2000\n",
    "start = time.time()\n",
    "\n",
    "runningloss = 0\n",
    "for it in range(10*epoch):\n",
    "    if it%10==0:\n",
    "        print(it)\n",
    "    \n",
    "    # Build batch\n",
    "    support, query, targs, _ = batchmaker(way,trainshot,testshot,trainset,alterful=True)\n",
    "    \n",
    "    # Predict\n",
    "    embed.zero_grad()\n",
    "    model.zero_grad()\n",
    "    embeds = embed(support)\n",
    "    qembeds = embed(query)\n",
    "    preds = model(embeds,qembeds,way)\n",
    "\n",
    "    # Calculate Loss\n",
    "    loss = criterion(preds, targs)\n",
    "    runningloss += loss.data[0]\n",
    "    \n",
    "    # Backprop\n",
    "    if it%epoch == 0:\n",
    "        optimizer = optim.Adam(embed.parameters(),lr=.001/(2**(it//epoch)))\n",
    "    loss.backward()\n",
    "#     nn.utils.clip_grad_norm(model.parameters(), 1)\n",
    "    optimizer.step()\n",
    "    \n",
    "    # Report\n",
    "    if it%vbity == vbity-1:\n",
    "        display.clear_output(wait=True)\n",
    "\n",
    "        losstracker.append(runningloss/vbity)\n",
    "        embed = embed.eval()\n",
    "        evalloss, evalacc, _ = evaluate(embed,model,criterion,evalway,trainshot,testshot,reps,testset)\n",
    "        embed = embed.train()\n",
    "        evallosstracker.append(evalloss)\n",
    "        evalacctracker.append(evalacc)\n",
    "        \n",
    "        pl.figure(1,figsize=(15,5))\n",
    "        pl.subplot(1,2,1)\n",
    "        pl.plot(losstracker)\n",
    "        pl.plot(evallosstracker)\n",
    "        pl.ylim((.5,3))\n",
    "        pl.title(\"Loss: Training Blue, Validation Gold\")\n",
    "        pl.subplot(1,2,2)\n",
    "        pl.plot(evalacctracker[::-1])\n",
    "        pl.ylim((0.3,.8))\n",
    "        pl.title(\"Validation Acc\")\n",
    "        pl.show()\n",
    "        \n",
    "        print(\"Train loss is: \"+str(runningloss/vbity)+\n",
    "              \"\\nValidation accuracy is: \"+str(evalacc)+\n",
    "              \"\\nValidation loss is: \"+str(evalloss)+\"\\n\")\n",
    "        runningloss = 0\n",
    "        print(time.time()-start)\n",
    "\n",
    "# MORE RECENT TEST\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(0.8505900807678699, 0.6548666666666668, 0.0091473772839506182)\n"
     ]
    }
   ],
   "source": [
    "criterion = nn.NLLLoss().cuda()\n",
    "embed = embed.eval()\n",
    "print(evaluate(embed,model,criterion,evalway,trainshot,testshot*3,200,testset))\n",
    "embed = embed.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.01318637251104336"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "1.96*np.sqrt(.0090525/200)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.65722400000000003"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.mean(evalacctracker[-50:])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/dww78/anaconda3/lib/python3.6/site-packages/torch/serialization.py:147: UserWarning: Couldn't retrieve source code for container of type ENCODER. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n",
      "/home/dww78/anaconda3/lib/python3.6/site-packages/torch/serialization.py:147: UserWarning: Couldn't retrieve source code for container of type Block. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n",
      "/home/dww78/anaconda3/lib/python3.6/site-packages/torch/serialization.py:147: UserWarning: Couldn't retrieve source code for container of type PREDICTOR. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n"
     ]
    }
   ],
   "source": [
    "name = 'naive_full'\n",
    "torch.save((embed,model,losstracker,evallosstracker,evalacctracker),'saved_models/'+name+'.pth')\n",
    "torch.save(embed.cpu().state_dict(),'saved_models/'+name+'_cpu.pth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "embed,model,losstracker,evallosstracker,evalacctracker = torch.load('saved_models/naive_4.pth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "Paper model achieves (over last 1000):\n",
    "    0.71156711111111126\n",
    "after 188500 iters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 213,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Variable containing:\n",
       "-6.7075e-03 -5.0079e+00 -2.7911e+01\n",
       "-2.9057e+00 -5.6263e-02 -4.5704e+01\n",
       "-1.8272e+01 -4.3870e+01 -1.1600e-08\n",
       "[torch.FloatTensor of size 3x3]"
      ]
     },
     "execution_count": 213,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "support = supportmaker(nclass,shot)\n",
    "inp = inpmaker(nclass)\n",
    "embeds = embed(support)\n",
    "query = embed(inp)\n",
    "model(embeds,query)"
   ]
  }
 ],
 "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}