PDL::Stats::Basic(3) basic statistics and related utilities such as standard deviation, Pearson correlation, and t-tests.

DESCRIPTION

The terms FUNCTIONS and METHODS are arbitrarily used to refer to methods that are threadable and methods that are NOT threadable, respectively.

Does not have mean or median function here. see SEE ALSO.

SYNOPSIS


use PDL::LiteF;
use PDL::NiceSlice;
use PDL::Stats::Basic;
my $stdv = $data->stdv;

or

    my $stdv = stdv( $data );

FUNCTIONS

stdv

  Signature: (a(n); float+ [o]b())

Sample standard deviation.

stdv processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

stdv_unbiased

  Signature: (a(n); float+ [o]b())

Unbiased estimate of population standard deviation.

stdv_unbiased processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

var

  Signature: (a(n); float+ [o]b())

Sample variance.

var processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

var_unbiased

  Signature: (a(n); float+ [o]b())

Unbiased estimate of population variance.

var_unbiased processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

se

  Signature: (a(n); float+ [o]b())

Standard error of the mean. Useful for calculating confidence intervals.

    # 95% confidence interval for samples with large N
    $ci_95_upper = $data->average + 1.96 * $data->se;
    $ci_95_lower = $data->average - 1.96 * $data->se;

se processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

ss

  Signature: (a(n); float+ [o]b())

Sum of squared deviations from the mean.

ss processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

skew

  Signature: (a(n); float+ [o]b())

Sample skewness, measure of asymmetry in data. skewness == 0 for normal distribution.

skew processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

skew_unbiased

  Signature: (a(n); float+ [o]b())

Unbiased estimate of population skewness. This is the number in GNumeric Descriptive Statistics.

skew_unbiased processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

kurt

  Signature: (a(n); float+ [o]b())

Sample kurtosis, measure of ``peakedness'' of data. kurtosis == 0 for normal distribution.

kurt processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

kurt_unbiased

  Signature: (a(n); float+ [o]b())

Unbiased estimate of population kurtosis. This is the number in GNumeric Descriptive Statistics.

kurt_unbiased processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

cov

  Signature: (a(n); b(n); float+ [o]c())

Sample covariance. see corr for ways to call

cov processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

cov_table

  Signature: (a(n,m); float+ [o]c(m,m))

Square covariance table. Gives the same result as threading using cov but it calculates only half the square, hence much faster. And it is easier to use with higher dimension pdls.

Usage:

    # 5 obs x 3 var, 2 such data tables
    perldl> $a = random 5, 3, 2
    perldl> p $cov = $a->cov_table
    [
     [
      [ 8.9636438 -1.8624472 -1.2416588]
      [-1.8624472  14.341514 -1.4245366]
      [-1.2416588 -1.4245366  9.8690655]
     ]
     [
      [   10.32644 -0.31311789 -0.95643674]
      [-0.31311789   15.051779  -7.2759577]
      [-0.95643674  -7.2759577   5.4465141]
     ]
    ]
    # diagonal elements of the cov table are the variances
    perldl> p $a->var
    [
     [ 8.9636438  14.341514  9.8690655]
     [  10.32644  15.051779  5.4465141]
    ]

for the same cov matrix table using cov,

    perldl> p $a->dummy(2)->cov($a->dummy(1))

cov_table processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

corr

  Signature: (a(n); b(n); float+ [o]c())

Pearson correlation coefficient. r = cov(X,Y) / (stdv(X) * stdv(Y)).

Usage:

    perldl> $a = random 5, 3
    perldl> $b = sequence 5,3
    perldl> p $a->corr($b)
    [0.20934208 0.30949881 0.26713007]

for square corr table

    perldl> p $a->corr($a->dummy(1))
    [
     [           1  -0.41995259 -0.029301192]
     [ -0.41995259            1  -0.61927619]
     [-0.029301192  -0.61927619            1]
    ]

but it is easier and faster to use corr_table.

corr processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

corr_table

  Signature: (a(n,m); float+ [o]c(m,m))

Square Pearson correlation table. Gives the same result as threading using corr but it calculates only half the square, hence much faster. And it is easier to use with higher dimension pdls.

Usage:

    # 5 obs x 3 var, 2 such data tables
 
    perldl> $a = random 5, 3, 2
    
    perldl> p $a->corr_table
    [
     [
     [          1 -0.69835951 -0.18549048]
     [-0.69835951           1  0.72481605]
     [-0.18549048  0.72481605           1]
    ]
    [
     [          1  0.82722569 -0.71779883]
     [ 0.82722569           1 -0.63938828]
     [-0.71779883 -0.63938828           1]
     ]
    ]

for the same result using corr,

    perldl> p $a->dummy(2)->corr($a->dummy(1))

This is also how to use t_corr and n_pair with such a table.

corr_table processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

t_corr

  Signature: (r(); n(); [o]t())

    $corr   = $data->corr( $data->dummy(1) );
    $n      = $data->n_pair( $data->dummy(1) );
    $t_corr = $corr->t_corr( $n );
    use PDL::GSL::CDF;
    $p_2tail = 2 * (1 - gsl_cdf_tdist_P( $t_corr->abs, $n-2 ));

t significance test for Pearson correlations.

t_corr processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

n_pair

  Signature: (a(n); b(n); indx [o]c())

Returns the number of good pairs between 2 lists. Useful with corr (esp. when bad values are involved)

n_pair processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

corr_dev

  Signature: (a(n); b(n); float+ [o]c())

    $corr = $a->dev_m->corr_dev($b->dev_m);

Calculates correlations from dev_m vals. Seems faster than doing corr from original vals when data pdl is big

corr_dev processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

t_test

  Signature: (a(n); b(m); float+ [o]t(); [o]d())

    my ($t, $df) = t_test( $pdl1, $pdl2 );
    use PDL::GSL::CDF;
    my $p_2tail = 2 * (1 - gsl_cdf_tdist_P( $t->abs, $df ));

Independent sample t-test, assuming equal var.

t_test processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

t_test_nev

  Signature: (a(n); b(m); float+ [o]t(); [o]d())

Independent sample t-test, NOT assuming equal var. ie Welch two sample t test. Df follows Welch-Satterthwaite equation instead of Satterthwaite (1946, as cited by Hays, 1994, 5th ed.). It matches GNumeric, which matches R.

    my ($t, $df) = $pdl1->t_test( $pdl2 );

t_test_nev processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

t_test_paired

  Signature: (a(n); b(n); float+ [o]t(); [o]d())

Paired sample t-test.

t_test_paired processes bad values. It will set the bad-value flag of all output piddles if the flag is set for any of the input piddles.

binomial_test

  Signature: (x(); n(); p_expected(); [o]p())

Binomial test. One-tailed significance test for two-outcome distribution. Given the number of successes, the number of trials, and the expected probability of success, returns the probability of getting this many or more successes.

This function does NOT currently support bad value in the number of successes.

Usage:

  # assume a fair coin, ie. 0.5 probablity of getting heads
  # test whether getting 8 heads out of 10 coin flips is unusual
  my $p = binomial_test( 8, 10, 0.5 );  # 0.0107421875. Yes it is unusual.

METHODS

rtable

Reads either file or file handle*. Returns observation x variable pdl and var and obs ids if specified. Ids in perl @ ref to allow for non-numeric ids. Other non-numeric entries are treated as missing, which are filled with $opt{MISSN} then set to BAD*. Can specify num of data rows to read from top but not arbitrary range.

*If passed handle, it will not be closed here.

*PDL::Bad::setvaltobad only works consistently with the default TYPE double before PDL-2.4.4_04.

Default options (case insensitive):

    V       => 1,        # verbose. prints simple status
    TYPE    => double,
    C_ID    => 1,        # boolean. file has col id.
    R_ID    => 1,        # boolean. file has row id.
    R_VAR   => 0,        # boolean. set to 1 if var in rows
    SEP     => "\t",     # can take regex qr//
    MISSN   => -999,     # this value treated as missing and set to BAD
    NROW    => '',       # set to read specified num of data rows

Usage:

Sample file diet.txt:

    uid height  weight  diet
    akw 72      320     1
    bcm 68      268     1
    clq 67      180     2
    dwm 70      200     2
  
    ($data, $idv, $ido) = rtable 'diet.txt';
    # By default prints out data info and @$idv index and element
    reading diet.txt for data and id... OK.
    data table as PDL dim o x v: PDL: Double D [4,3]
    0   height
    1   weight
    2   diet

Another way of using it,

    $data = rtable( \*STDIN, {TYPE=>long} );

group_by

Returns pdl reshaped according to the specified factor variable. Most useful when used in conjunction with other threading calculations such as average, stdv, etc. When the factor variable contains unequal number of cases in each level, the returned pdl is padded with bad values to fit the level with the most number of cases. This allows the subsequent calculation (average, stdv, etc) to return the correct results for each level.

Usage:

    # simple case with 1d pdl and equal number of n in each level of the factor
        pdl> p $a = sequence 10
        [0 1 2 3 4 5 6 7 8 9]
        pdl> p $factor = $a > 4
        [0 0 0 0 0 1 1 1 1 1]
        pdl> p $a->group_by( $factor )->average
        [2 7]
    # more complex case with threading and unequal number of n across levels in the factor
        pdl> p $a = sequence 10,2
        [
         [ 0  1  2  3  4  5  6  7  8  9]
         [10 11 12 13 14 15 16 17 18 19]
        ]
        pdl> p $factor = qsort $a( ,0) % 3
        [
         [0 0 0 0 1 1 1 2 2 2]
        ]
        pdl> p $a->group_by( $factor )
        [
         [
          [ 0  1  2  3]
          [10 11 12 13]
         ]
         [
          [  4   5   6 BAD]
          [ 14  15  16 BAD]
         ]
         [
          [  7   8   9 BAD]
          [ 17  18  19 BAD]
         ]
        ]
     ARRAY(0xa2a4e40)
    # group_by supports perl factors, multiple factors
    # returns factor labels in addition to pdl in array context
    pdl> p $a = sequence 12
    [0 1 2 3 4 5 6 7 8 9 10 11]
    pdl> $odd_even = [qw( e o e o e o e o e o e o )]
    pdl> $magnitude = [qw( l l l l l l h h h h h h )]
    pdl> ($a_grouped, $label) = $a->group_by( $odd_even, $magnitude )
    pdl> p $a_grouped
    [
     [
      [0 2 4]
      [1 3 5]
     ]
     [
      [ 6  8 10]
      [ 7  9 11]
     ]
    ]
    pdl> p Dumper $label
    $VAR1 = [
              [
                'e_l',
                'o_l'
              ],
              [
                'e_h',
                'o_h'
              ]
            ];

which_id

Lookup specified var (obs) ids in $idv ($ido) (see rtable) and return indices in $idv ($ido) as pdl if found. The indices are ordered by the specified subset. Useful for selecting data by var (obs) id.

    my $ind = which_id $ido, ['smith', 'summers', 'tesla'];
    my $data_subset = $data( $ind, );
    # take advantage of perl pattern matching
    # e.g. use data from people whose last name starts with s
    my $i = which_id $ido, [ grep { /^s/ } @$ido ];
    my $data_s = $data($i, );

REFERENCES

Hays, W.L. (1994). Statistics (5th ed.). Fort Worth, TX: Harcourt Brace College Publishers.

AUTHOR

Copyright (C) 2009 Maggie J. Xiong <maggiexyz users.sourceforge.net>

All rights reserved. There is no warranty. You are allowed to redistribute this software / documentation as described in the file COPYING in the PDL distribution.