Clustering on this reinforcement learning approach? -


i trying create agent selects action depending on state gives maximum reward.

to keep things simple keep 2 actions , 24 different states.

the states mimic hours in day , 2 actions web pages displayed user.

i still trying figure out how reward given , how policy depending on reward. plausible following:

between 0 , 1 determine 100% of probability. action should taken 1 chance of reward.

very simple example same state x:

if user shown page 1(action) , stays on (the action) reward due page one.

x = amount of rewards given state page 1 = 1 y = amount of rewards given state page 2 = 0

page 1 + page 2 = 1.0 chance x = ((x+y)/x) = 1/1 = 1.0 chance y = ((x+y)/y) = 1/0 = 0.0 1.0 chance page 1 correct action state 0.0 chance page 2 correct action state

the user shown page 1(action) due risk of reward being higher if displaying page 1 @ state. if user navigates instead page 2, page 2 reward.

x = amount of rewards given state page 1 = 1 y = amount of rewards given state page 2 = 1

page 1 + page 2 = 1.0 chance x = ((x+y)/x) = 2/1 = 0.5 chance y = ((x+y)/y) = 2/1 = 0.5 0.5 chance page 1 correct action state 0.5 chance page 2 correct action state

if user shown page 1(action) , stays on page 1, page 1 reward.

x = amount of rewards given state page 1 = 2 y = amount of rewards given state page 2 = 1

page 1 + page 2 = 1.0 chance x = 1.0 / ((x+y)*x) = 1.0 / 3*2 = 2/3 chance y = 1.0 / ((x+y)*x) = 1.0 / 3*1 = 1/3 2/3 chance page 1 correct action state 1/3 chance page 2 correct action state

as see updates , learns.

clustering

this work if days same, , know aren't. user might use page 1 week 1 , next week page 2 , week after page 1 , one. finding pattern needed somehow.

what trying achieve

i have following input data (state):

{     location: 'möllevångstorget, 21424, malmö',     weekday: 'monday',     time: '07:31' }

alternatively:

{     lat: 55.591538,     lon: 13.007153,     timestamp: '2015-03-03 07:31' }

or:

{     lat: 55.591538,     lon: 13.007153,     timestamp: 1427864271 // unix epoch time }

as can see, can manipulate inputs. it's important though include location , when occurred.

as mentioned before, finding patterns i'm worried about. wish predict when user going use application (displayed page), state created when user uses application.

another problem can see let's user uses application @ 07:30 1 week, uses 07:35 next , third week uses 07:32 around same location, algorithm should able determine user(environment) pick specific page(action).

basically predicting action user going choose.

i don’t think should use clustering. need implement function approximation. if have geo’s, reverse encode country. use country & city state inputs, e.g.: features might end being: is_america is_africa is_middle_east is_new_york is_morning is_afternoon etc

if country , city list grows big considering doing via relational reinforcement learning


Comments

Post a Comment

Popular posts from this blog

cakephp - simple blog with croogo -

i not problem, made simple blog in croogo, problem is not adding or modifying or delete, because arabic language ?? function add in controller `public function admin_add() { $this->set('title_for_layout', __('add part')); if (!empty($this->request->data)) { $this->part->create(); if ($this->part->save($this->request->data)) { $this->session->setflash(__('the part has been saved'), 'default', array('class' => 'success')); $this->redirect(array('action' => 'index'));//, $this->part->id)); } else { $this->session->setflash(__('the part not saved. please, try again.'), 'default', array('class' => 'error')); } } }` => model `class part extends appmodel { public $name = 'part'; var $belongst...
Read more

How to group boxplot outliers in gnuplot -

i have large set of data points. try plot them boxplot, of outliers exact same value , represented on line beside each other. found how set horizontal distance between outliers in gnuplot boxplot , doesn't much, apparently not possible. is possible group outliers together, print 1 point , print number in brackets beside indicate how many points there are? think make more readable in graph. for information, have 3 boxplots 1 x value , times 6 in 1 graph. using gnuplot 5 , played around pointsize, doesn't reduce distance anymore. hope can help! edit: set terminal pdf set output 'dat.pdf' file0 = 'dat1.dat' file1 = 'dat2.dat' file2 = 'dat3.dat' set pointsize 0.2 set notitle set xlabel 'x' set ylabel 'y' header = system('head -1 '.file0); n = words(header) set xtics ('' 1) set [i=1:n] xtics add (word(header, i) i) set style data boxplot plot file0 using (1-0.25):1:(0.2) boxplot lw 2 lc rgb '#8b0000...
Read more

bash - Performing variable substitution in a string -

i have string contains use of variable, , i'd substitute variable's value string. right best have is #!/bin/bash foo='$name; echo bar' name="the name" expanded="$(eval echo "$foo")" echo "$expanded" which has obvious defects: prints the name bar while i'd print the name; echo bar instead of eval can bash's regex matching bash's string replacement: foo='$name; echo bar' name="the name" [[ "$foo" =~ \$([[:alnum:]]+) ]] && s="${!bash_rematch[1]}" && expanded="${foo/\$${bash_rematch[1]}/$s}" echo "$expanded" name; echo bar
Read more