l_bfgs_b-dask-array.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Client: scheduler='tcp://localhost:8786' processes=7 cores=56>"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from dask import persist\n",
    "from dask.distributed import Client, progress\n",
    "client = Client('localhost:8786')\n",
    "client.restart()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import dask\n",
    "import dask.array as da\n",
    "import numpy as np\n",
    "from scipy.optimize import fmin_l_bfgs_b\n",
    "from sklearn import datasets\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "\n",
    "# make dataset\n",
    "X, y = datasets.make_classification(n_classes=2, n_samples=10000000)\n",
    "\n",
    "# create dask implementation\n",
    "Xda = da.from_array(X, 1000000)\n",
    "yda = da.from_array(y, 1000000)\n",
    "\n",
    "Xda, yda = dask.persist(Xda, yda)\n",
    "\n",
    "# logistic\n",
    "def sigmoid(x):\n",
    "    '''Sigmoid function of x.'''\n",
    "    return 1 / (1 + da.exp(-x))\n",
    "\n",
    "def compute_logistic_loss_grad(beta, X, y):\n",
    "    Xbeta = X.dot(beta)\n",
    "    # loss\n",
    "    eXbeta = da.exp(Xbeta)\n",
    "    loss_fn = (da.log1p(eXbeta)).sum() - da.dot(y, Xbeta)\n",
    "    # gradient\n",
    "    p = sigmoid(Xbeta)\n",
    "    gradient_fn = da.dot(X.T, p - y)\n",
    "    loss, gradient = dask.compute(loss_fn, gradient_fn)\n",
    "    return loss, gradient.copy()\n",
    "\n",
    "n, p = X.shape\n",
    "beta = np.zeros(p)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from distributed import wait"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# We balance the data around the cluster, then we duplicate it 10 times\n",
    "# Then we rebalance it again.  This is around 16GB\n",
    "\n",
    "wait([Xda, yda])\n",
    "client.rebalance([Xda, yda])\n",
    "\n",
    "Xda = da.concatenate([Xda] * 10, axis=0).persist()\n",
    "yda = da.concatenate([yda] * 10, axis=0).persist()\n",
    "\n",
    "wait([Xda, yda])\n",
    "client.rebalance([Xda, yda])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dask.array<concatenate, shape=(100000000, 20), dtype=float64, chunksize=(1000000, 20)>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Xda"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "16000000000"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Xda.nbytes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/anaconda/lib/python3.5/site-packages/dask/array/core.py:476: RuntimeWarning: overflow encountered in true_divide\n",
      "  o = func(*args, **kwargs)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 1.46 s, sys: 552 µs, total: 1.46 s\n",
      "Wall time: 9.98 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "with dask.set_options(fuse_ave_width=0):  # optimizations slows this down\n",
    "    new_beta_dask, loss_dask, info_dask = fmin_l_bfgs_b(\n",
    "        compute_logistic_loss_grad, beta, fprime=None,\n",
    "        args=(Xda, yda),\n",
    "        iprint=0, pgtol=1e-14, maxiter=10)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 25.6 s, sys: 2.24 s, total: 27.8 s\n",
      "Wall time: 14.1 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# sklearn LR with effectively no regularization and using scipy's L-BFGS optimizer\n",
    "# This is on the smaller, 1.6GB dataset\n",
    "lr = LogisticRegression(fit_intercept=False, solver=\"lbfgs\", C=1e5, max_iter=10)\n",
    "lr.fit(X, y)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ -7.18210521e-05   1.48211054e+00   8.59712361e-04  -5.36094181e-01\n",
      "   -4.73116035e-04   5.15202678e-04   7.14100165e-04  -7.67761494e-04\n",
      "   -2.20663555e-03   6.40228731e-01   2.15719078e-03   1.11574985e-03\n",
      "    3.01840479e-03  -8.91871490e-05  -1.20989408e-04  -7.35972279e-04\n",
      "   -2.26868882e-04  -2.03733770e-03  -1.12063873e+00  -4.72233441e-04]]\n",
      "[ -7.18210522e-05   1.48211054e+00   8.59712361e-04  -5.36094181e-01\n",
      "  -4.73116035e-04   5.15202679e-04   7.14100165e-04  -7.67761494e-04\n",
      "  -2.20663555e-03   6.40228731e-01   2.15719078e-03   1.11574985e-03\n",
      "   3.01840479e-03  -8.91871490e-05  -1.20989408e-04  -7.35972279e-04\n",
      "  -2.26868882e-04  -2.03733770e-03  -1.12063873e+00  -4.72233441e-04]\n"
     ]
    }
   ],
   "source": [
    "print(lr.coef_)\n",
    "print(new_beta_dask)\n"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda root]",
   "language": "python",
   "name": "conda-root-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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}