----------------------------------------------------------------------

-- This script implements the k-means algorithm.

--

-- The algorithm is implemented such that it produces filters, that

-- can be used in a convolutional network.

--

-- Eugenio Culurciello, Clement Farabet

----------------------------------------------------------------------

require 'image'

require 'unsup'

----------------------------------------------------------------------

-- parse command-line options

--

cmd = torch.CmdLine()

cmd:text()

cmd:text('Training a simple sparse coding dictionary on Berkeley images')

cmd:text()

cmd:text('Options')

cmd:option('-dir', 'outputs', 'subdirectory to save experiments in')

cmd:option('-datafile', 'http://torch7.s3-website-us-east-1.amazonaws.com/data/tr-berkeley-N5K-M56x56-lcn.ascii', 'Dataset URL')

cmd:option('-seed', 1, 'initial random seed')

cmd:option('-threads', 4, 'threads')

cmd:option('-inputsize', 9, 'size of each input patches')

cmd:option('-nkernels', 1024, 'number of kernels to learn')

cmd:option('-niter', 50, 'nb of k-means iterations')

cmd:option('-batchsize', 1000, 'batch size for k-means\' inner loop')

cmd:option('-nsamples', 500000, 'nb of random training samples')

cmd:option('-statinterval', 5000, 'interval for reporting stats/displaying stuff')

cmd:text()

params = cmd:parse(arg or {})

rundir = cmd:string('psd', params, {dir=true})

params.rundir = params.dir .. '/' .. rundir

if paths.dirp(params.rundir) then

os.execute('rm -r ' .. params.rundir)

end

os.execute('mkdir -p ' .. params.rundir)

cmd:addTime('psd')

cmd:log(params.rundir .. '/log.txt', params)

torch.manualSeed(params.seed)

torch.setnumthreads(params.threads)

----------------------------------------------------------------------

-- load data

--

dofile '1_data.lua'

filename = paths.basename(params.datafile)

if not paths.filep(filename) then

os.execute('wget ' .. params.datafile)

end

dataset = getdata(filename, params.inputsize)

dataset:conv()

----------------------------------------------------------------------

-- run k-means

--

-- create dataset

data = torch.Tensor(params.nsamples,params.inputsize*params.inputsize)

for i = 1,params.nsamples do

data[i] = dataset[i][1]

end

print '==> normalizing patches'

for i = 1,params.nsamples do

data[i]:add(-data[i]:mean())

data[i]:div(math.sqrt(data[i]:var() + 10))

end

print '==> whitening patches'

data = unsup.zca_whiten(data,nil,nil,nil,epsilon)

-- callback: display kernels

print '==> running k-means'

function cb (step,kernels)

local filters = {}

local ks = kernels:reshape(params.nkernels,1,params.inputsize,params.inputsize)

for i = 1,params.nkernels do

filters[i] = ks[i]:clone():div(ks[i]:max())

end

if itorch then

itorch.image(filters)

end

end

-- run k-means

kernels = unsup.kmeans(data, params.nkernels, params.niter, params.batchsize, cb, true)

-- save kernels

file = 'kmeans_'..params.nkernels..'.t7'

print('==> saving centroids to disk: ' .. file)

torch.save(file, kernels)