adding math functions with a machine learning package

master
Steve L. Nyemba 10 years ago
parent 39ff5c7e1b
commit 501e043dd8

@ -0,0 +1,254 @@
/**
* Simple Javascript eXtension - 1.0
* (c) 2011 - 2015 Steve L. Nyemba, steve@the-phi.com
* License GPL version 3.0
*
* dependencies:
* utils.js implementation of design patterns and utilities
*
* This file contains an enhancement of utilities integrated into the jx.math.* built-in package of javascript
* Because we implement math and numerical functions it is to be understood that most of the functions will have common preconditions
* i.e lxi.constructor == Array && isNumber(lxi) unless specified otherwise
* jx.math.max
* jx.math.min
* jx.math.sum
* jx.math.prod
* jx.math.freq
* jx.math.avg
* jx.math.mean computes the mean/average of a list of observations (arthmetic mean included too)
* jx.math.sd computes the standard deviation of a list of observations
* jx.math.var computes the variance of a list of observations
* jx.math.diff computes the absolute difference of values in an array
* jx.math.fibonacci comptutes the fibonacci value of a given number
* jx.math.factorial computes the factorial of a given number
*/
if(!jx){
var jx = {} ;
}
jx.math = {}
jx.math.sqrt = Math.sqrt;
jx.math.PHI = (1+jx.math.sqrt(5))/2 ;//1.61803399 ;
/**
* @param lxi list of observatins xi
*/
jx.math.max = function(lxi){
sortNumber= function(a,b) {
return a - b;
}
index = lxi.length -1 ;
max = jx.utils.cast(lxi,Number).sort(sortNumber)[index] ; // perhaps need to cast
return max ;
}
/**
* finds the minimum of a list of observation lxi (vector of values)
* @param lxi list/vector of values/observations
*/
jx.math.min = function(lxi){
sortNumber = function(a,b){
return a- b;
}
min = jx.utils.cast(lxi,Number).sort(sortNumber)[0] ;
return min ;
}
/**
* @pre : values.constructor == Array
* @param lxi list of observed values to be summed
*/
jx.math.sum = function(lxi){
return eval(lxi.join('+'));
} ;
/**
* This function will compute the frequency of a vector i.e providing relative values
*/
jx.math.freq = function(lxi){
var N = jx.math.sum(lxi) ;
return jx.utils.patterns.visitor(lxi,function(xi){
return xi/N ;
});
}
/**
* This function will perform the product of a vector
* @pre lxi.constructor == Array && isNumber(lxi)
*/
jx.math.prod = function(lxi){
return eval(lxi.join('*')) ;
}
/**
* @pre : lni != null && lxi.length == lni.length
* @param lxi list of observed values
* @param lni list of the number of times observations of index i have been made
*/
jx.math.avg = function(lxi,lni){
N = lxi.length ;
if(lni == null){
return jx.math.sum(lxi)/N ;
}else{
values = []
for(var i=0; i < lxi.length; i++){
values[i] = Number(lxi[i])*Number(lni[i]) ;
}
return Number(jx.math.sum(values)/N) ;
}
};
jx.math.mean = jx.math.avg ;
jx.math.pow = Math.pow
jx.math.sd = function(lxi,lni){
N = lxi.length ;
mean = jx.math.mean(lxi,lni) ;
sqr = [] ;
for(var i=0; i < lxi.length ;i++){
console.log(lxi[i])
sqr[i] = jx.math.pow((Number(lxi[i])-mean),2 ) ;
}
total = jx.math.sum(sqr);
return jx.math.sqrt(total/(N-1)) ;
} ;
/**
* Computes the factorial of a given value
*/
jx.math.factorial = function(value){
r =value;
for(var i =value-1; i > 0; i--){
r *= i ;
}
return r;
} ;
/**
* Computes the fibonacci value of a given number using the golden ratio
*/
jx.math.fibonacci = function(value){
r = (jx.math.pow(jx.math.PHI,value)/jx.math.sqrt(5)) + 0.5 ;
return jx.math.floor(r) ;
} ;
/**
* computes the absolute difference of values in a list of observations
*/
jx.math.diff = function(lxi){
var r = [] ;
var x,y;
for(var i=0; i < lxi.length-1; i++){
x = lxi[i] ;
y = lxi[i+1] ;
r.push(y-x)
}
return r ;
};
/**
* This section implements a few handlers based on sets
*/
jx.math.sets = {} ;
/**
* This function will perform a unique operation of values/objects
* @param list list/vector of values or objects
* @param equals operator to be used, only provide this for complex objects
*/
jx.math.sets.unique = jx.utils.unique ;
/**
* This function will perform the union of 2 sets (objects, or values)
* @param list1 list/vector of values or objects
* @param list2 list/vector of values or objects
* @param equals operator to be used to evaluate equality (use this for complex objects)
*/
jx.math.sets.union = function(list1,list2,equals){
runion = [] ;
runion = list1.concat(list2) ;
runion = jx.math.sets.unique(runion,equals)
return runion;
}
/**
* This function will normalize values within a vector
* By definition normalization is (x - u) / sd (assuming population parameters are known)
*/
jx.math.normalize = function(lvalues){
mean = jx.math.mean(lvalues) ;
sd = jx.math.sd(lvalues) ;
return jx.utils.patterns.visitor(lvalues,function(x){
return ((x - mean) / sd)
})
}
/**
* This function will scale a feature vector over it's range
*/
jx.math.scale = function(lvalues,percent){
max = jx.math.max(lvalues) ;
min = jx.math.min(lvalues) ;
return jx.utils.patterns.visitor(lvalues,function(x){
var value = (x - min ) / max ;
if(percent == true){
return (100*value).toFixed(2)
}else{
return value ;
}
})
}
/**
* This is a lightweight map reduce infrastructure
*/
jx.mr = {} ;
/**
* This function will perform a map on a given id in rec, then will call emit with the
*/
jx.mr.map = null
/**
* @param keys
* @param values array of values that were mapped
*/
jx.mr.reduce = null;
jx.mr.mapreduce = function(data,fn_map,fn_reduce){
if (fn_map == null){
throw new "Map function is not defined"
}
map = {} ;
emit = function(id,values){
if(map[id] == null){
map[id] = []
}
map[id].push(values);
}
if(data.constructor != Array){
for (id in data){
//rec = data[id] ;
rec = {}
rec['__id'] = id;
rec['data'] = data[id] ;
fn_map(rec,emit)
}
}else{
for (var i=0; i < data.length; i++){
rec = data[i];
fn_map(rec,emit);
//if(i == 2)break;
}
}
if(fn_reduce != null){
keys = jx.utils.keys(map) ;
m = {}
for(var i=0; i < keys.length; i++){
id = keys[i] ;
values = map[id] ;
value = fn_reduce(id,values) ;
id = keys[i] ;
m[id] = value;
}
map = m
}
return map ;
}

@ -0,0 +1,61 @@
/**
* Simple Javascript eXtension - 1.0, Machine Leanring Module
* (c) 2011 - 2015 Steve L. Nyemba, steve@the-phi.com
* License GPL version 3.0
*
* dependencies:
* jx.utils collection of utilities and design patterns used
* jx.math various math & statistical functions
* This file implements a few reusable machine learning models/techniques
*
* jx.ml.mapreduce Performs a standard/basic mapreduce (single thread for now)
* jx.ml.regression Will perform linear & logistic regressions
*/
*
if(!jx){
var jx = {} ;
}
jx.ml = {}
/**
* The function performs map/reduce and at the very least map i.e the reduce function is optional
*/
jx.ml.mapreduce = function(data,fn_map,fn_reduce){
//
// insure that the mapping function has been provided
//
var __map = {}
var emit = function(id,mvalue){
if(__map[id] == null){
__map[id] = []
}
__map[id].push(mvalue) ;
}//-- end of the emitter
if(data.constructor != Array){
jx.utils.patterns.visitor(data,function(id){
fn_map(data[id],emit) ;
});
}else{
jx.utils.patterns.visitor(data,function(i){
fn_map(data[i],emit) ;
});
}
if(fn_reduce != null){
//
// We will be performing a reduce operation at this point
var ids = jx.utils.keys(__map) ;
jx.utils.patterns.visitor(ids,function(id){
return __map[id] = fn_reduce(id,__map[id]) ;
});
}
return __map ;
}//--
/**
* The modules developed below will perform linear regression and logistic regression
*/
jx.ml.regression = {}

@ -1,6 +1,6 @@
/**
* Simple Javascript eXtension - 1.0
* (c) 2014 - 2015 Steve L. Nyemba, steve@the-phi.com
* (c) 2011 - 2015 Steve L. Nyemba, steve@the-phi.com
* License GPL version 3.0
*
* Implementation of miscellaneous utilities commonly used, These functions are reusable and simple:

Loading…
Cancel
Save