Hybrid_custom_model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from torchvision.models import resnet18\n",
    "from torchvision import transforms\n",
    "import numpy as np\n",
    "import torch\n",
    "import torch.autograd as autograd\n",
    "from torch.autograd import Variable\n",
    "import torch.nn as nn\n",
    "import pylab as pl\n",
    "import time\n",
    "import pyinn.ncrelu as ncrelu\n",
    "import math\n",
    "from IPython import display\n",
    "import torch.nn.functional as F\n",
    "torch.cuda.set_device(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50 550\n"
     ]
    }
   ],
   "source": [
    "training_data = torch.load('/data/mini_imagenet/mIN_96.pth')\n",
    "testing_data = [i[-50:] for i in training_data]\n",
    "training_data = [i[:-50] for i in training_data]\n",
    "print(len(testing_data[0]),len(training_data[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from gpytorch.utils.fft import fft3,ifft3\n",
    "from torch.autograd import Variable, Function\n",
    "from torch.nn.parameter import Parameter\n",
    "\n",
    "class FFT3(Function):\n",
    "    def __init__(self):\n",
    "        super(FFT3, self).__init__()\n",
    "    \n",
    "    def forward(self, inp):\n",
    "        return fft3(inp)\n",
    "    \n",
    "    def backward(self, grad):\n",
    "        return ifft3(grad)\n",
    "    \n",
    "class FFT3_up(Function):\n",
    "    def __init__(self):\n",
    "        super(FFT3_up, self).__init__()\n",
    "    \n",
    "    def forward(self, inp):\n",
    "        result = fft3(inp)\n",
    "        return result\n",
    "    \n",
    "    def backward(self, grad):\n",
    "        return ifft3(grad).sum(0).unsqueeze(0)\n",
    "    \n",
    "class iFFT3(Function):\n",
    "    def __init__(self):\n",
    "        super(iFFT3, self).__init__()\n",
    "        \n",
    "    def forward(self, inp):\n",
    "        return ifft3(inp)\n",
    "    \n",
    "    def backward(self, grad):\n",
    "        return fft3(grad)\n",
    "    \n",
    "class Conv3dF(nn.Module):\n",
    "    def __init__(self,nchannels,size,padding=False):\n",
    "        super(Conv3dF,self).__init__()\n",
    "        self.nchannels = nchannels\n",
    "        self.fft = FFT3()\n",
    "        self.ifft = iFFT3()\n",
    "        self.fftup = FFT3_up()\n",
    "        self.padding = padding\n",
    "        self.striper = Variable(torch.bernoulli(torch.ones(1,nchannels,1,1)/2)*2-1, requires_grad=False).cuda()\n",
    "        self.padder = (nn.ConstantPad2d((0,size-3+2,0,size-3+2),0) if padding else \n",
    "                       nn.ConstantPad2d((0,size-3,0,size-3),0))\n",
    "        self.weights = Parameter(torch.randn(1,nchannels,3,3)*2/math.sqrt(nchannels)) \n",
    "                                                # TODO proper weight initialization\n",
    "        \n",
    "    def forward(self, inp):\n",
    "        fi = self.fft(F.pad(self.striper*inp,(1,1,1,1))) if self.padding else self.fft(self.striper*inp)\n",
    "        fw = self.fftup(self.padder(self.weights))\n",
    "        result = torch.cat([(fi[:,:,:,:,0]*fw[:,:,:,:,0]-fi[:,:,:,:,1]*fw[:,:,:,:,1]).unsqueeze(4),\n",
    "                            (fi[:,:,:,:,0]*fw[:,:,:,:,1]+fi[:,:,:,:,1]*fw[:,:,:,:,0]).unsqueeze(4)],dim=4)\n",
    "#         print(\"GOTHERE fft elem mult\")\n",
    "        return self.ifft(result)[:,:,1:-1,1:-1].contiguous() if self.padding else self.ifft(result)\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "class fftBlock(nn.Module):\n",
    "    def __init__(self, size, imsize, padding=False):\n",
    "        super(fftBlock, self).__init__()\n",
    "        self.layers = nn.Sequential(\n",
    "            Conv3dF(size, imsize, padding),\n",
    "            nn.BatchNorm2d(size)\n",
    "        )\n",
    "        \n",
    "    def forward(self, inp):\n",
    "#         print(\"GOTHERE fft block\",inp.size())\n",
    "        return ncrelu(self.layers(inp))\n",
    "    \n",
    "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",
    "        )\n",
    "        \n",
    "    def forward(self, inp):\n",
    "#         print(\"GOTHERE block\",inp.size())\n",
    "        return ncrelu(self.layers(inp))\n",
    "\n",
    "class ENCODER(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(ENCODER, self).__init__()\n",
    "        self.process = nn.Sequential(\n",
    "            nn.Conv2d(3,64,kernel_size=3,padding=1),\n",
    "            Block(64,64),\n",
    "            nn.MaxPool2d(2),\n",
    "            Block(128,128),\n",
    "            nn.MaxPool2d(2),\n",
    "            Block(256,256),\n",
    "            nn.MaxPool2d(2),\n",
    "            fftBlock(512,12,padding=True),\n",
    "            nn.MaxPool2d(2),\n",
    "            fftBlock(1024,6,padding=True),\n",
    "            nn.AvgPool2d((6,6)))\n",
    "        self.final = nn.Sequential(\n",
    "            nn.Linear(2048,100)\n",
    "        )\n",
    "            \n",
    "    def forward(self, inp):\n",
    "        out = self.process(inp)\n",
    "        return self.final(out.view(-1, 2048))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2.029795287489891, 0.4926)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.eval()\n",
    "evaluate(model, criterion, testing_data, shot=50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "standardize = transforms.Compose([\n",
    "            transforms.ToTensor(),\n",
    "            transforms.Normalize(mean=[.485,.456,.406],std=[.229,.224,.225])\n",
    "            ])\n",
    "\n",
    "alter = transforms.Compose([\n",
    "            transforms.ToPILImage(),\n",
    "            transforms.RandomHorizontalFlip(),\n",
    "            transforms.RandomCrop(80,padding=10),\n",
    "            standardize\n",
    "            ])\n",
    "\n",
    "def batchmaker(theset, way=20, shot=1, alterful=False):\n",
    "    classes = np.random.choice(len(theset),way)\n",
    "    if alterful:\n",
    "        li = [torch.cat([alter(theset[cl][i]).view(1,3,80,80) for i in \n",
    "                         np.random.choice(len(theset[0]),shot)],dim=0).float()\n",
    "              for cl in classes]\n",
    "    else:\n",
    "        li = [torch.cat([standardize(theset[cl][i]).view(1,3,80,80) for i in \n",
    "                         np.random.choice(len(theset[0]),shot)],dim=0).float()\n",
    "              for cl in classes]\n",
    "    support = torch.cat([t[:shot,:,:,:] for t in li],dim=0)\n",
    "    stargs = torch.LongTensor(np.repeat(classes, shot))\n",
    "#     query = torch.cat([t[trainshot:,:,:,:] for t in li],dim=0)\n",
    "#     qtargs = torch.LongTensor([i//shot for i in range(testshot*way)])\n",
    "    return(Variable(support, volatile=(not alterful)).cuda(),\n",
    "           Variable(stargs, volatile=(not alterful)).cuda()\n",
    "          )\n",
    "# batchmaker(training_data,shot=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "criterion = nn.CrossEntropyLoss().cuda()\n",
    "# optimizer = torch.optim.Adam(model.parameters())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def evaluate(model, criterion, testing_data, shot=10, way=100):\n",
    "    support, targs = batchmaker(testing_data, way=way, shot=shot)\n",
    "    assert (shot*way)%20==0\n",
    "    batches = int(shot*way/20)\n",
    "    support = support.view(batches, 20, 3, 80, 80)\n",
    "    targs = targs.view(batches, 20)\n",
    "    acc = 0\n",
    "    loss = 0\n",
    "    for i in range(batches):\n",
    "        preds = model(support[i,:,:,:,:])\n",
    "        loss += criterion(preds, targs[i,:]).data[0]\n",
    "        _,bins = torch.max(preds, 1)\n",
    "        acc += torch.sum(torch.eq(bins,targs[i,:])).data[0]\n",
    "    acc = float(acc)/shot/way\n",
    "    loss = float(loss)/batches\n",
    "    return loss, acc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "999076  parameters in neural net.\n"
     ]
    }
   ],
   "source": [
    "model = ENCODER()\n",
    "model.cuda()\n",
    "nweights = sum([i.numel() for i in list(model.parameters())])\n",
    "print(nweights,\" parameters in neural net.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA20AAAE/CAYAAADVKysfAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f21d486ca58>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train loss is: 1.9121760475635527\n",
      "Validation accuracy is: 0.47600000000000003\n",
      "Validation loss is: 2.0721118104457856\n",
      "\n",
      "2008.0359811782837\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "vbity = 200\n",
    "epoch = 5000\n",
    "start = time.time()\n",
    "losstracker = []\n",
    "evalacctracker = []\n",
    "evallosstracker = []\n",
    "runningloss = 0\n",
    "limit = 6\n",
    "for it in range(limit*epoch):\n",
    "    if it%50==0:\n",
    "        print(it)\n",
    "    \n",
    "    # Build batch\n",
    "    support, targs = batchmaker(training_data, alterful=True)\n",
    "    \n",
    "    # Predict\n",
    "    model.zero_grad()\n",
    "    preds = model(support)\n",
    "\n",
    "    # Calculate Loss\n",
    "    loss = criterion(preds, targs)\n",
    "    runningloss += loss.data[0]\n",
    "    \n",
    "    # Backprop\n",
    "    if it%epoch == 0:\n",
    "        optimizer = torch.optim.Adam(model.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",
    "        model = model.eval()\n",
    "        evalloss, evalacc = evaluate(model, criterion, testing_data, shot=20)\n",
    "        model = model.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((0,4))\n",
    "#         pl.xlim((0,limit*epoch))\n",
    "        pl.title(\"Loss: Training Blue, Validation Gold\")\n",
    "        pl.subplot(1,2,2)\n",
    "        pl.plot(evalacctracker[::-1])\n",
    "        pl.ylim((0,1))\n",
    "#         pl.xlim((0,limit*epoch))\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)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.47816999999999998"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "name = \"hybrid_1024\"\n",
    "torch.save((embed,model,losstracker,evallosstracker,evalacctracker),'saved_models/'+name+'.pth')"
   ]
  },
  {
   "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": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}