Different results with Python and command line

[jtom12]

Problem: Hello there,

I'm obtaining different results depending on how I create the model, whether using Python or the command line with the executable.
The issue seems to be related to border set similarity. To support my explanation, I have studied the following four cases:

• One feature trained without borders: Get the same set.
• One feature trained with borders: Get the same set.
• Two features trained without borders: Getthe same set.
• Two features trained with borders: Get different sets.

I have attached a simple script with an example for each of the four cases mentioned above.

catboost version: catboost-1.2.3 (Same version for Python and command line)
Operating System: Ubuntu 22.04.4 LTS
CPU: 12th Gen Intel® Core™ i5-12500H × 16
GPU: None

[ek-ak]

Hello!
Thank you for you issue! We will investigate the problem.