Measures.Quartiles: value for Q1 (lower quartile) wrong in QuantileMethod.R

[hkoestin]

Hello @cesarsouza

I just found out, that the calculation for the lower quartile (25% quantile) is not working correctly, when passing in QuantileMethod.R as method.

I have the following vector:

var v1 = new double[] { 18, 31, 25, 2, 22, 13, 37, 1, 4, 7, 6, 45, 10, 24, 23, 49, 27, 9, 35, 
                             14, 34, 33, 41, 42, 20, 43, 3, 48, 15, 39, 11, 38, 46, 17, 40, 16, 
                             50, 29, 19, 47, 12, 28, 32, 8, 30, 26, 5, 44, 36, 21};

For this vector, I get the following results for the Q1 value:

• v1:
  • using Accord: 12.5
  • using R: 13.25

Don't know though, where the difference is coming from.
For some other samples I get correct results.

Cheers, Harald

[cesarsouza]

Hi @hkoestin,

Thanks for opening the issue! As a workaround, you can sort the values before calling the Quartiles method while also passing alreadySorted as false, as shown below:

double m = Measures.Quartiles(v1.Sorted(), out q1, out q3, alreadySorted: true, type: QuantileMethod.R);

I am still investigating where the problem actually is, but I hope the workaround can solve the issue for you in the meantime!

Thanks again,
Cesar

[cesarsouza]

By the way, regarding the second vector, I am getting a value of 9.25 in R instead of the indicated 11.25. This matches with the framework once the values have been sorted manually with the workaround above. So I hope the workaround will really work for you, but if you find another case that it fails, please let me know.

Regards,
Cesar

[hkoestin]

@cesarsouza 👍
Thanks for the quick response. I double-checked v2 again. And in fact, it is really 9.25 I get from R and Accord. So no worries about that. I updated the initial issue description.

I will include the fix for sorting the values first. Thanks again!

[hkoestin]

Did the fix with the code suggestion from above. Works fine now.

If you want to or need them, here is my set of test-vectors for the regression tests against R-compatibility, used to execute xUnit Theories.

First double[] is the input vector.
Second value is Q1.
Third value is Q3.

new object[]
{
    new double?[] {0.0, 1.0, 2.0, 4.0, 5.4, 3.5, 7.8, 8.9, 17.0, 23.78, 98.9, 2.3, 4.5, 6.7, 9.34, 42.42},
    3.2, 11.255
},
new object[]
{
    new double?[] {0.0, 5.4, 2.0, 4.0, 1.0, 3.5, 7.8, 17.0, 8.9, 98.9, 23.78, 2.3, 4.5, 6.7, 42.42, 9.34},
    3.2, 11.255
},
new object[]
{
    new double?[]
    {
        0.0, 1.0, 2.0, 4.0, 5.4, 3.5, 7.8, 8.9, 17.0, 23.78, 98.9, 2.3, 4.5, 6.7, 9.34, 42.42, 23, 17.87, 18.54, 16.23, 15.34,
        19.8723, 23, 24.32
    },
    4.375, 20.65423
},
new object[]
{
    new double?[]
    {
        18, 31, 25, 2, 22, 13, 37, 1, 4, 7, 6, 45, 10, 24, 23, 49, 27, 9, 35, 14, 34, 33, 41, 42, 20, 43, 3, 48, 15, 39, 11, 38,
        46, 17, 40, 16, 50, 29, 19, 47, 12, 28, 32, 8, 30, 26, 5, 44, 36, 21
    },
    13.25, 37.75
},
new object[]
{
    new double?[]
    {
        18, 14, 1, 15, 4, 32, 10, 26, 38, 9, 24, 16, 31, 20, 25, 30, 22, 6, 28, 21, 33, 17, 5, 35, 2, 13, 36, 8, 29, 7
    },
    9.25, 28.75
},
new object[]
{
    new double?[]
    {
        112, 718, 320, 576, 547, 658, 253, 560, 408, 314, 681, 303, 236, 753, 122, 239, 222, 797, 593, 274, 338, 604, 52, 245,
        389
    },
    245, 593
},
new object[]
{
    new double?[]
    {
        209, 556, 317, 571, 219, 599, 568, 516, 582, 279, 298, 319, 614, 290, 458, 262, 281, 606, 513, 519, 356, 338, 525, 576,
        180
    },
    290, 568
}

[cesarsouza]

Thanks @hkoestin!

I've committed a fix to the issue yesterday and it has passed all test vectors above. The fix should be included in the next (pre-)release.

Regards,
Cesar

[cesarsouza]

Added in 3.8.0.