search

PicNet / XGBoost.Net

.Net wrappers for the awesome XGBoost library
51 stars 18 forks source link

Inconsistent learning results with python's xgboost #27

Closed libu1129 closed 5 years ago

libu1129 commented 5 years ago

I have moved the code written in python xgboost to c # xgboost.net I did the same with the same data, but the results were different. I have this problem because of the float-type decimal length problem, It is expected that the data entered in python is different from the data entered in decimal length, resulting in different results. Is there any possibility that I expected to happen? I am trying to test this by changing float type to double type to solve this problem.

This article is written in Google Translator and may be difficult to read. summary : I put the same data as python's xgboost, but got different results. Thank you.

libu1129 commented 5 years ago

The float type was not a problem. Even if the data consist of 0 and 1, the results are different.

libu1129 commented 5 years ago

I realized that existing python source and data preprocessing parts are partly different. I modified that part so I could get the same result.