distributions-truncated-normal-pdf

Truncated normal distribution probability density function (PDF)

Downloads in past

Stats

StarsIssuesVersionUpdatedCreatedSize
distributions-truncated-normal-pdf
100.0.08 years ago8 years agoMinified + gzip package size for distributions-truncated-normal-pdf in KB

Readme

Probability Density Function
!NPM versionnpm-imagenpm-url !Build Statustravis-imagetravis-url !Coverage Statuscodecov-imagecodecov-url !Dependenciesdependencies-imagedependencies-url
Truncated normaltruncated-normal distribution probability density function (PDF).

The distribution of a normally distributed random variable X conditional on a < X < b is a truncated normaltruncated-normal distribution. The probability density functiondensity-function (PDF) for a truncated normaltruncated-normal random variable is
<img src="https://cdn.rawgit.com/distributions-io/truncated-normal-pdf/420be8917ad8ff8b29d3f21f88744581bbff9445/docs/img/eqn.svg" alt="Probability density function (PDF) for a truncated normal distribution.">
<br>

where Phi and phi denote the cumulative distribution functioncdf and density functiondensity-function of the normalnormal distribution, respectively, mu is the location and sigma > 0 is the scale parameter of the distribution. a and b are the minimum and maximum support.

Installation

$ npm install distributions-truncated-normal-pdf

For use in the browser, use browserify.

Usage

var pdf = require( 'distributions-truncated-normal-pdf' );

pdf( x, options )

Evaluates the probability density functiondensity-function (PDF) for the truncated normaltruncated-normal distribution. x may be either a number, an array, a typed array, or a matrix.
var matrix = require( 'dstructs-matrix' ),
	mat,
	out,
	x,
	i;

out = pdf( 1 );
// returns 0.242

out = pdf( -1 );
// returns 0.242

x = [ 0, 0.5, 1, 1.5, 2, 2.5 ];
out = pdf( x );
// returns [ ~0.399, ~0.352, ~0.242, 0.13, ~0.054, ~0.018 ]

x = new Float32Array( x );
out = pdf( x );
// returns Float64Array( [~0.399,~0.352,~0.242,0.13,~0.054,~0.018] )

x = new Float64Array( 6 );
for ( i = 0; i < 6; i++ ) {
	x[ i ] = i*0.5;
}
mat = matrix( x, [3,2], 'float64' );
/*
	[ 0 0.5
	  1 1.5
	  2 2.5 ]
*/

out = pdf( mat );
/*
	[ ~0.399 ~0.352
	  ~0.242 0.13
	  ~0.054 ~0.018 ]
*/

The function accepts the following options:
  • a: minimum support. Default: -Infinity
  • b: maximum support. Default: +Infinity
  • mu: location parameter. Default: 0.
  • sigma: scale parameter. Default: 1.
  • accessor: accessor function for accessing array values.
  • dtype: output typed array or matrix data type. Default: float64.
  • copy: boolean indicating if the function should return a new data structure. Default: true.
  • path: deepget/deepset key path.
  • sep: deepget/deepset key path separator. Default: '.'.

A truncated normaltruncated-normal distribution is a function of four parameters: a and b, the minimum and maximum support, mu(location parameter) and sigma > 0(scale parameter). By default, a = -Infinity and b = +Infinity, mu is equal to 0 and sigma is equal to 1. To adjust either parameter, set the corresponding option.
var x = [ 0, 0.5, 1, 1.5, 2, 2.5 ];

var out = pdf( x, {
	'a': -5,
	'b': 5,
	'mu': 2,
	'sigma': 2,
});
// returns [ 0.13, ~0.161, ~0.189, ~0.207, ~0.214, ~0.207 ]

For non-numeric arrays, provide an accessor function for accessing array values.
var data = [
	[0,0],
	[1,0.5],
	[2,1],
	[3,1.5],
	[4,2],
	[5,2.5]
];

function getValue( d, i ) {
	return d[ 1 ];
}

var out = pdf( data, {
	'accessor': getValue
});
// returns [ ~0.399, ~0.352, ~0.242, 0.13, ~0.054, ~0.018 ]

To deepset an object array, provide a key path and, optionally, a key path separator.
var data = [
	{'x':[0,0]},
	{'x':[1,0.5]},
	{'x':[2,1]},
	{'x':[3,1.5]},
	{'x':[4,2]},
	{'x':[5,2.5]}
];

var out = pdf( data, {
	'path': 'x/1',
	'sep': '/'
});
/*
	[
		{'x':[0,~0.399]},
		{'x':[1,~0.352]},
		{'x':[2,~0.242]},
		{'x':[3,0.13]},
		{'x':[4,~0.054]},
		{'x':[5,~0.018]}
	]
*/
var bool = ( data === out );
// returns true

By default, when provided a typed array or matrix, the output data structure is float64 in order to preserve precision. To specify a different data type, set the dtype option (see matrix for a list of acceptable data types).
var x, out;

x = new Int8Array( [0,1,2,3,4] );

out = pdf( x, {
	'mu': 2,
	'sigma': 2,
	'dtype': 'int32'
});
// returns Int32Array( [0,0,0,0,0] )

// Works for plain arrays, as well...
out = pdf( [0,0.5,1,1.5,2], {
	'mu': 2,
	'sigma': 2,
	'dtype': 'uint8'
});
// returns Uint8Array( [0,0,0,0,0] )

By default, the function returns a new data structure. To mutate the input data structure (e.g., when input values can be discarded or when optimizing memory usage), set the copy option to false.
var bool,
	mat,
	out,
	x,
	i;

x = [ 0, 0.5, 1, 1.5, 2 ];

out = pdf( x, {
	'copy': false
});
// returns [ ~0.399, ~0.352, ~0.242, 0.13, ~0.054 ]

bool = ( x === out );
// returns true

x = new Int16Array( 6 );
for ( i = 0; i < 6; i++ ) {
	x[ i ] = i*0.5;
}
mat = matrix( x, [3,2], 'float32' );
/*
	[ 0 0
	  1 1
	  2 2 ]
*/

out = pdf( mat, {
	'copy': false
});
/*
	[ ~0.399 ~0.399
	  ~0.242 ~0.242
	  ~0.054 ~0.054 ]
*/

bool = ( mat === out );
// returns true

Notes

  • If an element is not a numeric value, the evaluated PDF is NaN.
``` javascript
var data, out;

out = pdf( null );
// returns NaN

out = pdf( true );
// returns NaN

out = pdf( {'a':'b'} );
// returns NaN

out = pdf( [ true, null, [] ] );
// returns [ NaN, NaN, NaN ]

function getValue( d, i ) {
	return d.x;
}
data = [
	{'x':true},
	{'x':[]},
	{'x':{}},
	{'x':null}
];

out = pdf( data, {
	'accessor': getValue
});
// returns [ NaN, NaN, NaN, NaN ]

out = pdf( data, {
	'path': 'x'
});
/*
	[
		{'x':NaN},
		{'x':NaN},
		{'x':NaN,
		{'x':NaN}
	]
*/
```
  • Be careful when providing a data structure which contains non-numeric elements and specifying an integer output data type, as NaN values are cast to 0.
``` javascript
var out = pdf( [ true, null, [] ], {
	'dtype': 'int8'
});
// returns Int8Array( [0,0,0] );
```

Examples

var pdf = require( 'distributions-truncated-normal-pdf' ),
	matrix = require( 'dstructs-matrix' );

var data,
	mat,
	out,
	tmp,
	i;

// Plain arrays...
data = new Array( 10 );
for ( i = 0; i < data.length; i++ ) {
	data[ i ] = -2.5 + i * 0.5;
}
out = pdf( data );

// Object arrays (accessors)...
function getValue( d ) {
	return d.x;
}
for ( i = 0; i < data.length; i++ ) {
	data[ i ] = {
		'x': data[ i ]
	};
}
out = pdf( data, {
	'accessor': getValue
});

// Deep set arrays...
for ( i = 0; i < data.length; i++ ) {
	data[ i ] = {
		'x': [ i, data[ i ].x ]
	};
}
out = pdf( data, {
	'path': 'x/1',
	'sep': '/'
});

// Typed arrays...
data = new Float32Array( 10 );
for ( i = 0; i < data.length; i++ ) {
	data[ i ] = -2.5 + i * 0.5;
}
out = pdf( data );

// Matrices...
mat = matrix( data, [5,2], 'float32' );
out = pdf( mat );

// Matrices (custom output data type)...
out = pdf( mat, {
	'dtype': 'uint8'
});

To run the example code from the top-level application directory,
$ node ./examples/index.js

Tests

Unit

Unit tests use the Mocha test framework with Chai assertions. To run the tests, execute the following command in the top-level application directory:
$ make test

All new feature development should have corresponding unit tests to validate correct functionality.

Test Coverage

This repository uses Istanbul as its code coverage tool. To generate a test coverage report, execute the following command in the top-level application directory:
$ make test-cov

Istanbul creates a ./reports/coverage directory. To access an HTML version of the report,
$ make view-cov

License

MIT license.

Copyright

Copyright © 2016. The Compute.io Authors.