Randomly changing results from predictions #2530
-
|
Hi, I have a model saved that seems to (generally) have parameters that are expected/explainable. I am loading this model with I'm using Splink=4.0.5 and Python=3.12.3. |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 1 reply
-
|
Hmmm, that is puzzling. Nothing immediately leaps to mind as the likely cause - if you are not retraining the model then Any further details you are able to provide (the One thing that might help narrow it down is to try and investigate the record pairs that are not always appearing - if you save the left+right ids of your |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for getting back to me so quickly! It appears that the randomness is coming from a data manipulation step in Polars. I was able to get repeatable results using Pandas end to end or using a saved, post-manipulation dataset. I will look into what it is in Polars that's causing the randomness. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the update, glad you resolved the problem |
Beta Was this translation helpful? Give feedback.
Hmmm, that is puzzling. Nothing immediately leaps to mind as the likely cause - if you are not retraining the model then
predict()should give the same set of results consistently for the same input data.Any further details you are able to provide (the
.jsonfile, or your comparisons, or the script you are running, backend) might point to some clues. Are there any possibly non-deterministic functions in your comparisons (or blocking rules)? And is the input data definitely unchanged between these runs?One thing that might help narrow it down is to try and investigate the record pairs that are not always appearing - if you save the left+right ids of your
df_predicttable to file you can …