libsvm model file format

According to this FAQ the model format in libsvm should be straightforward. And in fact it is, when I call just svm-train. As an example, the first SV for the a1a dataset is

 1 3:1 11:1 14:1 19:1 39:1 42:1 55:1 64:1 67:1 73:1 75:1 76:1 80:1 83:1

On the other hand, if I use the easy.py script, my first SV ends up being:

 512 1:-1 2:-1 3:1 4:-1 5:-1 6:-1 7:-1 8:-1 9:-1 10:-1 11:1 13:-1 14:1 15:-1 16:-1 17:-1 18:-1 19:1 20:-1 21:-1 22:-1 23:-1 24:-1 25:-1 26:-1 27:-1 28:-1 29:-1 30:-1 31:-1 32:-1 33:-1 34:-1 35:-1 36:-1 37:-1 38:-1 39:1 40:-1 41:-1 42:1 43:-1 44:-1 45:-1 46:-1 47:-1 48:-1 49:-1 50:-1 51:-1 52:-1 53:-1 54:-1 55:1 56:-1 57:-1 58:-1 59:-1 61:-1 62:-1 63:-1 64:1 65:-1 66:-1 67:1 68:-1 69:-1 70:-1 71:-1 72:-1 73:1 74:-1 75:1 76:1 77:-1 78:-1 79:-1 80:1 81:-1 82:-1 83:1 84:-1 85:-1 86:-1 87:-1 88:-1 90:-1 91:-1 92:-1 93:-1 94:-1 95:-1 97:-1 98:-1 99:-1 100:-1 101:-1 102:-1 103:-1 104:-1 105:-1 106:-1 107:-1 108:-1 109:-1 110:-1 112:-1 113:-1 114:-1 115:-1 117:-1 118:-1 119:-1 

which is an instance that doesn't exist at all in my training set! In fact if I do:

 $ grep "119:" a1a
 -1 1:1 6:1 18:1 22:1 36:1 42:1 49:1 66:1 67:1 73:1 74:1 76:1 80:1 119:1 
 -1 1:1 6:1 18:1 26:1 35:1 43:1 53:1 65:1 67:1 73:1 74:1 76:1 80:1 119:1 
 -1 2:1 6:1 15:1 19:1 39:1 42:1 55:1 62:1 67:1 72:1 74:1 76:1 78:1 119:1 
 -1 4:1 6:1 16:1 21:1 35:1 44:1 49:1 64:1 67:1 72:1 74:1 76:1 78:1 119:1 
 -1 2:1 6:1 14:1 30:1 35:1 42:1 49:1 65:1 67:1 72:1 74:1 76:1 78:1 119:1 
 -1 2:1 6:1 17:1 20:1 37:1 40:1 57:1 63:1 67:1 73:1 74:1 76:1 80:1 119:1 
 -1 5:1 6:1 18:1 22:1 36:1 40:1 54:1 61:1 67:1 72:1 75:1 76:1 80:1 119:1 
 -1 5:1 6:1 17:1 26:1 35:1 42:1 53:1 62:1 67:1 73:1 74:1 76:1 80:1 119:1 

There isn't any instance with 119:-1 (and even if it's just swapping +1 with -1, there isn't any instance with 119:1 and 118:1 either - missing attributes are zeros)

If I do this source code modification, I clearly see that in the former case (only svm-train involved) the first SV is also the first instance. But in the latter case (i.e. with easy.py s开发者_运维问答cript), the output which should give me which instance is the SV is eaten by grid.py

What's going on, here?

---

I think the culprit here is probably the call easy.py makes to svm-scale, which scales each attribute to be within [-1,1]. The training examples sent to svm-train will not be the same ones that are in your training file.