TCEP dataset incoherent with 'official' version?

[ArnoVel]

Hi, After I opened the issue about the labels being all set to 1, I went to check the tcep reference website to identify some pairs that got permuted and so on.

I stumbled across something strange: in the original dataset, some of the variables are multivariate. This can be seen, for example, in pair 54 or pair 71.

However, in the current version of cdt, when checking these two pairs, one finds 1D variables.

> data.iloc[53]
A    [43.51, 41.33, 36.78, -8.82, 34.61, 40.11, 12....
B    [42.0, 75.0, 69.0, 42.0, 76.0, 72.0, 77.0, 81....
Name: pair54, dtype: object
> data.iloc[53]['A']
array([ 43.51,  41.33,  36.78,  -8.82,  34.61,  40.11,  12.52, -35.18,
        48.12,  40.24,  25.4 ,  26.19,  23.71,  13.09,  53.97,  50.83,
        17.25,   6.48,  27.44, -16.3 ,  43.86, -24.66, -15.78,   4.94,
        42.71,  12.35,  -3.38,  11.54,   3.86,  45.42,   4.36,  12.11,
        49.42, -33.45,  31.14, -11.7 ,  -4.25,  -4.33,   9.92,   5.33,
        45.8 ,  23.  ,  35.17,  50.08,  55.68,  11.58,  18.48,  -0.23,
        30.06,  13.7 ,   3.75,  15.33,  59.43,   9.  ,   6.92, -18.14,
        60.17,  48.86,   4.93, -17.54,   0.39,  13.44,  41.7 ,  52.52,
         5.54,  37.97,  12.05,  16.  ,  13.47,  14.62,   9.54,  11.86,
         6.8 ,  18.54,  14.08,  22.3 ,  47.5 ,  64.14,  28.63,  -6.19,
        35.71,  33.32,  53.34,  31.79,  41.9 ,  18.  ,  35.68,  31.94,
        51.18,  -1.28,  39.02,  37.51,  29.37,  42.87,  17.97,  56.95,
        33.89, -29.3 ,   6.31,  32.88,  54.69,  49.61,  22.18,  42.  ,
       -18.92, -13.99,   3.15,   4.17,  12.65,  35.9 ,  14.6 ,  18.07,
       -20.16,  19.42,  47.91,  42.46,  33.99, -25.97,  19.74, -22.57,
        27.71,  52.37,  12.1 , -22.28, -41.29,  12.15,  13.52,   9.06,
        59.91,  23.61,  33.68,  31.88,   8.99,  -9.47, -25.3 , -12.09,
        14.58,  52.22,  38.71,  18.45,  25.29,  47.01, -20.87,  44.45,
        55.76,  27.15,  14.  , -13.83,   0.34,  24.67,  14.7 ,  44.8 ,
         8.47,   1.29,  48.21,  46.05,  -9.43,   2.04, -25.75,  40.42,
         6.92,  13.2 ,  15.63,   5.82, -26.32,  59.33,  46.95,  33.52,
        38.57,  -6.17,  13.76,  -8.57,   6.12, -21.14,  10.66,  36.81,
        39.94,  37.95,   0.31,  50.44,  24.48,  51.5 ,  38.89,  18.34,
       -34.89,  41.32, -17.74,  10.5 ,  21.03,  15.36, -15.41, -17.82])

Same can be seen about pair 71. Is this a mistake, or just a shuffling of the data? I made sure I set shuffle=False before testing for the two pairs. If the basic (non-shuffled) dataset is already shuffled, or has been pre-processed in some way to reduce dimensionality, can we have some explanation of how the two datasets relate to each other?

Any amount of information would help, Thanks

[diviyank]

Hi, This is concerning, I will look at how I got this version of the TCEP and come back to you.

Best, Diviyan

[ArnoVel]

Hi, Would it be possible to have some kind of update? Thanks!

[ArnoVel]

Hi, I checked the official website in details, It highly likely that the current CDT TCEP version is simply the current TCEP (with 108 pairs) with the multivariate one removed. This is likely as it leaves 99 pairs, which is the current length of the CDT TCEP. However I did not take the time to check if the two match. Regards, A.V

[diviyank]

Hi, Sorry for the delay, I was quite busy lately. Thanks for looking into it; it seems to be indeed the case. I just checked all the pairs and they match. I will add another dataset containing all the pairs (including the multivariate ones) because most of the algorithms do not support multivariate variables. Best regards, Diviyan