home | news | discuss | issues | license LURE: sk |
|
---|---|
Scott Knott testtim@menzies.us Recursive bi-clustering of set of |
require "show"
local the=require "config"
local tiles=require "tiles"
local lst=require "lists"
|
Simple utils |
local function nth(t,n) return t._all[ math.floor(#t._all*n) ] end
local function mid(t) return nth(t,0.5) end
local function iqr(t) return nth(t,0.75) - nth(t,0.25) end
|
Here's a simple counter, used to track |
local function create() return {
_all={}, sum=0,n=0, mu=0} end
|
Here how to update a counter with one value |
local function update(i,x)
i._all[#i._all+1]=x
i.sum = i.sum + x
i.n = i.n + 1
i.mu = i.sum/i.n end
|
Here how to update a counter with many values from |
local function updates(t, counter)
counter = counter or create()
for j=1,#t do
update(counter,t[j]) end
return counter end
|
This code is always counting left and right within the list of samples. To save time, memo all those "peeks". |
local function memo(samples,here,stop,_memo, b4,inc)
if stop > here then inc=1 else inc=-1 end
if here ~= stop then
b4= lst.copy( memo(samples,here+inc, stop, _memo)) end
_memo[here] = updates(samples[here]._all, b4)
return _memo[here] end
|
Seek a split that maximizes the expected value of the square of the difference in means before and after the split. At that point, if the two splits are not statistically the same, then recurse into each part of the split. |
return function (samples,epsilon,same)
epsilon = epsilon or the.sample.epsilon
local function split(lo,hi,all,rank,lvl)
local best,lmemo,rmemo = 0,{},{}
memo(samples,hi,lo, lmemo) -- summarize i+1 using i
memo(samples,lo,hi, rmemo) -- summarize i using i+1
local cut, lbest, rbest
for j=lo,hi-1 do -- step1: look for the best cut
local l = lmemo[j]
local r = rmemo[j+1]
if mid(l)*the.sample.epsilon < mid(r) then
if not same(l,r) then
local tmp= l.n/all.n*(l.mu - all.mu)^2 +
r.n/all.n*(r.mu - all.mu)^2
if tmp > best then
cut = j
best = tmp
lbest = lst.copy(l)
rbest = lst.copy(r) end end end end
if cut then -- step2a: use the cut (if you found it)
rank = split(lo, cut, lbest, rank, lvl+1) + 1
rank = split(cut+1, hi, rbest, rank, lvl+1)
else -- step2b: otherwise, all samples get same rank
for j=lo,hi do
samples[j].rank = rank end end
return rank
end
table.sort(samples, function (x,y) return
mid(x) < mid(y) end)
split(1,#samples, memo(samples,1,#samples,{}),1,0)
return samples end
|
LegalLURE, Copyright (c) 2017, Tim Menzies All rights reserved, BSD 3-Clause License Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
|