exploration.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    }
   ],
   "source": [
    "%pylab inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# common imports\n",
    "from __future__ import print_function\n",
    "import os\n",
    "import pickle\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import scipy.ndimage as ndi\n",
    "import sklearn\n",
    "import tensorflow as tf\n",
    "import torch\n",
    "from matplotlib import gridspec\n",
    "from scipy.stats import kde\n",
    "from sklearn import decomposition\n",
    "from torch import nn, optim\n",
    "#from automlp import automlp\n",
    "from automlp import automlp\n",
    "\n",
    "import ipdb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from automlp import automlp, parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_model():\n",
    "    return nn.Sequential(nn.Linear(28*28, 100),\n",
    "                         nn.ReLU(),\n",
    "                         nn.Linear(100, 10))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchvision.datasets import mnist\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "from torchvision import transforms, utils\n",
    "\n",
    "class MnistAdapter(object):\n",
    "    def __init__(self, ds):\n",
    "        self.ds = ds\n",
    "    def __len__(self):\n",
    "        return len(self.ds)\n",
    "    def __getitem__(self, index):\n",
    "        inputs, targets = self.ds[index]\n",
    "        return torch.FloatTensor(np.array(inputs)/255.0).reshape(-1), int(targets)\n",
    "\n",
    "data = MnistAdapter(mnist.MNIST(\"./_data\", download=True))\n",
    "test_data = MnistAdapter(mnist.MNIST(\"./_data\", download=True, train=False))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "ps = parameters.ParameterSet()\n",
    "ps += parameters.LogParameter(\"lr\", 1e-5, 1e1)\n",
    "ps += parameters.QuantizedLogParameter(\"bs\", 5, 500)\n",
    "ex = parameters.Exploration(ps)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 0.0989 OrderedDict([('lr', 1.718630436028197e-05), ('bs', 38)])\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-10-677258c5f33b>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      5\u001b[0m     \u001b[0mtrainer\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mautomlp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTrainer\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"crossentropy\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      6\u001b[0m     \u001b[0;32mfor\u001b[0m \u001b[0mepoch\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m30\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 7\u001b[0;31m         \u001b[0mtrainer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtrain_dataset\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0moptions\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mpv\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mntrain\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m100000\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      8\u001b[0m         \u001b[0mloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtrainer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mevaluate_classification\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtest_data\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      9\u001b[0m     \u001b[0;32mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtrial\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mloss\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mpv\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/work/automlp/automlp.py\u001b[0m in \u001b[0;36mtrain_dataset\u001b[0;34m(self, dataset, ntrain, batch_size, options)\u001b[0m\n\u001b[1;32m    230\u001b[0m         \u001b[0mlosses\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[1;32m    231\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mntrain\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m//\u001b[0m\u001b[0mint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_size\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[0;32m--> 232\u001b[0;31m             \u001b[0msample\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mgetbatch_random\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch_size\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    233\u001b[0m             \u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtargets\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmake_batch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msample\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmake_batch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msample\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    234\u001b[0m             \u001b[0mloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtrain_batch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtargets\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/work/automlp/automlp.py\u001b[0m in \u001b[0;36mgetbatch_random\u001b[0;34m(dataset, batch_size, convert)\u001b[0m\n\u001b[1;32m    104\u001b[0m \u001b[0;32mdef\u001b[0m \u001b[0mgetbatch_random\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch_size\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mconvert\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mfrom_numpy\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m     \u001b[0mn\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m     \u001b[0mbatch\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mrandint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_size\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    107\u001b[0m     \u001b[0minputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mconvert\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msample\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0msample\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mbatch\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m     \u001b[0mtargets\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mconvert\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msample\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0msample\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mbatch\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m<ipython-input-5-032315498f16>\u001b[0m in \u001b[0;36m__getitem__\u001b[0;34m(self, index)\u001b[0m\n\u001b[1;32m     10\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m__getitem__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mindex\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     11\u001b[0m         \u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtargets\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mds\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mindex\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 12\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mFloatTensor\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[0minputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m/\u001b[0m\u001b[0;36m255.0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtargets\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     13\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mMnistAdapter\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmnist\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mMNIST\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"./_data\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdownload\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "for trial in range(200):\n",
    "    pv = ex.pick_farthest()\n",
    "    model = make_model()\n",
    "    model.cuda()\n",
    "    trainer = automlp.Trainer(model, \"crossentropy\")\n",
    "    for epoch in range(30):\n",
    "        trainer.train_dataset(data, options=pv, ntrain=3000000)\n",
    "        loss = trainer.evaluate_classification(test_data)\n",
    "    print(trial, loss, pv)\n",
    "    ex.add_result(pv, loss)\n",
    "    model.cpu()\n",
    "    with open(\"mnist-100-sgd-{}.pth\", \"wb\") as stream:\n",
    "        torch.save(model, stream)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "xscale('log'); yscale('log')\n",
    "s = array([np.array(r[1].values()) for r in ex.results])\n",
    "scatter(s[:,0], s[:,1], c=[r[0] for r in ex.results])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "min(ex.results)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "reload(parameters)\n",
    "reload(automlp)\n",
    "ps = parameters.ParameterSet()\n",
    "ps += parameters.LogParameter(\"lr\", 1e-5, 1e1)\n",
    "ps += parameters.LogParameter(\"beta1\", 1e-3, 3e-1)\n",
    "ps += parameters.LogParameter(\"beta2\", 1e-4, 3e-2)\n",
    "ps += parameters.QuantizedLogParameter(\"bs\", 20, 200)\n",
    "ps += parameters.Constant(\"eps\", 1e-8)\n",
    "ex = parameters.Exploration(ps)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for trial in range(200):\n",
    "    pv = ex.pick_farthest()\n",
    "    model = make_model()\n",
    "    trainer = automlp.Trainer(model, \"crossentropy\")\n",
    "    model.gpu()\n",
    "    for epoch in range(30):\n",
    "        trainer.train_dataset(data, options=options, ntrain=3000000)\n",
    "        loss = trainer.evaluate_classification(test_data)\n",
    "        print(i, loss, pv)\n",
    "    ex.add_result(pv, loss)\n",
    "    model.cpu()\n",
    "    with open(\"mnist-100-adam-{}.pth\", \"wb\") as stream:\n",
    "        torch.save(model, stream)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "figsize(12,6)\n",
    "s = array([np.array(r[1].values()) for r in ex.results])\n",
    "subplot(121); xscale('log'); yscale('log'); scatter(s[:,0], s[:,1], c=[clip(r[0], 0, 0.03) for r in ex.results])\n",
    "subplot(122); xscale('log'); yscale('log'); scatter(s[:,1], s[:,2], c=[clip(r[0], 0, 0.03) for r in ex.results])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "min(ex.results)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}