nvidia-AI.summerschool17-no.../teaching-lw/126-linear-lsq.ipynb

238 lines
156 KiB
Text
Raw Permalink Normal View History

2017-09-22 17:26:46 +00:00
{
"metadata": {
"name": "111-linear-lsq"
},
"nbformat": 3,
"nbformat_minor": 0,
"worksheets": [
{
"cells": [
{
"cell_type": "code",
"collapsed": true,
"input": [
"import scipy,pylab,numpy\n",
"from numpy import *\n",
"from scipy import *\n",
"from pylab import *\n",
"from urllib import urlopen\n",
"from gzip import GzipFile\n",
"from scipy.spatial import distance\n",
"from scipy.spatial.distance import cdist"
],
"language": "python",
"metadata": {},
"outputs": [],
"prompt_number": 1
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Linear Classification Problem\n",
"===============\n",
"\n",
"Let's start by generating a simple linear classification problem. That is, a problem in which the optimal decision boundary is known to be linear. We're also generating a problem in which the samples are perfectly linearly separable. This is the exception rather than the rule."
]
},
{
"cell_type": "code",
"collapsed": false,
"input": [
"data = random_sample((1000,2))\n",
"labels = (data[:,0]*0.7+data[:,1]*0.4>0.5)\n",
"sum(labels)"
],
"language": "python",
"metadata": {},
"outputs": [
{
"output_type": "pyout",
"prompt_number": 1,
"text": [
"548"
]
}
],
"prompt_number": 1
},
{
"cell_type": "code",
"collapsed": false,
"input": [
"d0 = data[labels==False]\n",
"d1 = data[labels]\n",
"figure(figsize=(6,6)); xlim((0,1)); ylim((0,1))\n",
"plot(d0[:,0],d0[:,1],\"bo\")\n",
"plot(d1[:,0],d1[:,1],\"ro\")\n",
"print len(d0),len(d1)"
],
"language": "python",
"metadata": {},
"outputs": [
{
"output_type": "stream",
"stream": "stdout",
"text": [
"452 548"
]
},
{
"output_type": "display_data",
"png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAFtCAYAAAD8jSoDAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nShUEPzl9lZ1j0fUiqwbQOBGk
}
],
"prompt_number": 2
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"As before, we *augment* the vectors to get rid of the inhomogeneous part."
]
},
{
"cell_type": "code",
"collapsed": false,
"input": [
"augmented = concatenate([ones((len(data),1)),data],axis=1)\n",
"augmented[:3,:]"
],
"language": "python",
"metadata": {},
"outputs": [
{
"output_type": "pyout",
"prompt_number": 3,
"text": [
"array([[ 1. , 0.35265835, 0.82892312],\n",
" [ 1. , 0.09204218, 0.87157948],\n",
" [ 1. , 0.83583516, 0.77753499]])"
]
}
],
"prompt_number": 3
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"It's convenient to use labels of (-1,1) instead of (0,1); that way, the weight vector we get out of least squares is comparable to the weight vector we're getting out of the perceptron learning algorithm."
]
},
{
"cell_type": "code",
"collapsed": true,
"input": [
"labels = 2.0*labels-1.0"
],
"language": "python",
"metadata": {},
"outputs": [],
"prompt_number": 4
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Least Square Solution\n",
"====================\n",
"\n",
"We now want to find a vector $a$ such that $a \\cdot x_i \\approx c_i$ for all $i$.\n",
"\n",
"We can write this as a matrix and vector equation if we put the measurements into the rows of a matrix $X = {x_1\\choose x_N}$ and $c = {c_1\\choose c_N}$\n",
"\n",
"$$X \\cdot a^T = c$$\n",
"\n",
"In order to solve this, we take the pseudo-inverse of $X$:\n",
"\n",
"$$a^T = X^\\dagger \\cdot c$$\n",
"\n",
"The pseudo-inverse is a generalization of the inverse matrix for non-rectangular and/or non-full-rank matrices."
]
},
{
"cell_type": "code",
"collapsed": false,
"input": [
"a = dot(linalg.pinv(augmented),labels)\n",
"d,a0,a1 = a\n",
"print d,a0,a1"
],
"language": "python",
"metadata": {},
"outputs": [
{
"output_type": "stream",
"stream": "stdout",
"text": [
"-1.81688247454 2.62589047604 1.2659712297"
]
}
],
"prompt_number": 5
},
{
"cell_type": "code",
"collapsed": false,
"input": [
"d0 = data[labels<=0]\n",
"d1 = data[labels>0]\n",
"figure(figsize=(6,6)); xlim((0,1)); ylim((0,1))\n",
"plot(d0[:,0],d0[:,1],\"bo\")\n",
"plot(d1[:,0],d1[:,1],\"ro\")\n",
"plot([0,-d/a0],[-d/a1,0],\"g\")\n",
"savefig(\"tmp.png\")\n",
"print len(d0),len(d1)"
],
"language": "python",
"metadata": {},
"outputs": [
{
"output_type": "stream",
"stream": "stdout",
"text": [
"452 548"
]
},
{
"output_type": "display_data",
"png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAFtCAYAAAD8jSoDAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nuHv1aq4+igBtnrbk5qKCskBJ
}
],
"prompt_number": 6
},
{
"cell_type": "code",
"collapsed": false,
"input": [
"xs = linspace(0.0,1.0,100)[:,newaxis]\n",
"ys = linspace(0.0,1.0,100)[newaxis,:]\n",
"xs = xs + 0*ys\n",
"ys = ys + 0*xs\n",
"image = xs*a0+ys*a1+d\n",
"imshow(image.T[::-1]); savefig(\"temp.png\")"
],
"language": "python",
"metadata": {},
"outputs": [
{
"output_type": "display_data",
"png": "iVBORw0KGgoAAAANSUhEUgAAAP0AAAD8CAYAAAC8aaJZAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nmQta2KWsSMYDmnmm3nR6z4P5
}
],
"prompt_number": 7
},
{
"cell_type": "code",
"collapsed": true,
"input": [],
"language": "python",
"metadata": {},
"outputs": []
}
],
"metadata": {}
}
]
}