The output of lime text is not consistent with rise text

[SarahAlidoost]

For text, both methods RISE and LIME return a tuple of (token, index, importance), however, the output of RISE is sorted by the "index" whereas they are sorted by "importance" in LIME's output. We can sort the values in lime in the function _reshape_result_for_single_label.

RISE example output:

[('A', 0, 0.7009708379209041),
 ('delectable', 1, 0.8650634929537773),
 ('and', 2, 0.7003263755142688),
 ('intriguing', 3, 1.0261991372704504),
 ('thriller', 4, 0.9036030551791191),
 ('filled', 5, 0.6704552842676639),
 ('with', 6, 0.7485938167572022),
 ('surprises', 7, 0.7003605794906616),
 ('and', 8, 0.7706552624702453)]

LIME example output

[('intriguing', 3, 0.15290728964252595),
 ('delectable', 1, 0.06392747194386154),
 ('thriller', 4, 0.06170429812738322),
 ('A', 0, 0.02412706125892299),
 ('and', 2, 0.021543642932292974),
 ('filled', 5, -0.013505093974254312),
 ('with', 6, 0.010936157339420459),
 ('surprises', 7, 0.0031371849149267623)]

[SarahAlidoost]

As discussed in standup, we decided to sort the output of the lime method by index.