Minor flaws in model export
IMO there are several minor issues in the JSON export of the model:
"total_chain_models" can be better
- trained_model_path contains the absolute path. Is it have a reason fronm the reproducibility point of view (since that path becomes non-actual if the folder is moved)? Maybe the relative path is better?
- The "nodes_from" indices are duplicated if there is only one parent for node
- The FileExistsError occurs if the model exports twice with the same file name. The warning and additional index in the file name look comfortable behavior for the user (now the results of the long optimization run can be lost in such case).