PiecewiseLinearDistribution cannot compute `mean`, `std`, and `var` when truncated.

[pr0m1th3as]

There is some issue with mean, std, and var methods in truncated PiecewiseLinearDistribution objects. It seems to be related to integral built-in function throwing an error.

>> pkg load statistics
>> load patients
>> [f, x] = ecdf (Weight);
>> f = f(1:5:end);
>> x = x(1:5:end);
>> pd = PiecewiseLinearDistribution (x, f);
>> t = truncate (pd, 130, 180)
t =
  PiecewiseLinearDistribution

F(111) = 0
F(118) = 0.05
F(124) = 0.13
F(130) = 0.25
F(135) = 0.37
F(142) = 0.5
F(163) = 0.55
F(171) = 0.61
F(178) = 0.7
F(183) = 0.82
F(189) = 0.94
F(202) = 1
  Truncated to the interval [130, 180]

>> mean (pd)
ans = 153.61
>> mean (t)
error: quadcc: integrand F must return a single, real-valued vector of the same size as the input
error: called from
    integral at line 139 column 11
    mean at line 243 column 11
>>

[NRJank]

Integral requires the integrand functions to be vectorized such that it can pass an array of inputs and get an array of outputs. (Parallel processing the quadrature routine.). There is an "ArrayValued" option that will generally let you work with non vectorized functions, but there will be a performance penalty

[pr0m1th3as]

Just pushed a change to address this issue. I don't don't have access to MATLAB at the moment. Can you confirm that the expected values are also true in MATLAB, before I close this issue?

load patients
[f, x] = ecdf (Weight);
f = f(1:5:end);
x = x(1:5:end);
pd = PiecewiseLinearDistribution (x, f);
t = truncate (pd, 130, 180);
mean (t)    # it must be 152.311
std(t)         # it must be 18.2941
var(t)         # it must be 334.6757

[pr0m1th3as]

closing as fixed