You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi all, I am currently using pykeen to train a RotatE model. My dataset currently looks like this:
ID
Feature1
Feature2
Feature3
Output
ID1
verylow_1
high_2
verylow_3
Output1
ID2
low_1
high_2
verylow_3
Output2
ID3
low_1
high_2
high_3
Output3
ID4
low_1
high_2
low_3
Output4
ID5
veryhigh_1
low_2
low_3
Output5
ID6
veryhigh_1
reasonable_2
verylow_3
Output5
ID7
low_1
reasonable_2
high_3
Output6
ID8
high_1
reasonable_2
high_3
Output7
ID9
high_1
reasonable_2
high_3
Output7
ID10
high_1
high_2
high_3
Output8
ID11
high_1
high_2
low_3
Output9
ID12
high_1
low_2
high_3
Output10
ID13
high_1
low_2
veryhigh_3
Output11
ID14
high_1
low_2
veryhigh_3
Output11
ID15
high_1
high_2
verylow_3
Output12
ID16
veryhigh_1
reasonable_2
Output13
ID17
high_1
Output13
ID18
veryhigh_1
low_2
Output13
ID19
low_1
low_2
veryhigh_3
Output14
ID20
low_1
very low_2
veryhigh_3
Output15
ID21
high_2
low_3
with the triplets being, e.g. (ID1, hasFeature1, verylow_1), (ID1, hasFeature2, high_2), etc.
I am trying to do a tail prediction on ID21, so (ID21, hasOutput, ?)
Based on the precedence information, the combination of high_2 and low_3 should give Output4 or Output9 as the result. However, the model fails to predict this - instead giving Output5 half the time, with a random spattering of other predictions - so this is not accurate enough for my use, almost as if it's selecting it randomly.
I've tried other things, like embedding the ID into the tail feature as well, e.g. (ID1, hasFeature1, ID1_verylow_1) but this makes it seemingly more random instead.
Is there something I can do to make this more accurate and consistent? Like using a different model, etc.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
Hi all, I am currently using pykeen to train a RotatE model. My dataset currently looks like this:
with the triplets being, e.g. (ID1, hasFeature1, verylow_1), (ID1, hasFeature2, high_2), etc.
I am trying to do a tail prediction on ID21, so (ID21, hasOutput, ?)
Based on the precedence information, the combination of high_2 and low_3 should give Output4 or Output9 as the result. However, the model fails to predict this - instead giving Output5 half the time, with a random spattering of other predictions - so this is not accurate enough for my use, almost as if it's selecting it randomly.
I've tried other things, like embedding the ID into the tail feature as well, e.g. (ID1, hasFeature1, ID1_verylow_1) but this makes it seemingly more random instead.
Is there something I can do to make this more accurate and consistent? Like using a different model, etc.
Beta Was this translation helpful? Give feedback.
All reactions