Catboost: set_params method is not working

[abdullahalsaidi16]

Problem: I'm trying to change the number of iterations after I load the model weights but this not changing anything
catboost version: 0.17
Operating System: ubuntu 18.04
Code:
model = CatBoostRegressor()
model.load_model('boost.cbm')
model = model.set_params(iterations=10)
print(model.get_all_params())

Output:
{'nan_mode': 'Min', 'eval_metric': 'RMSE', 'combinations_ctr': ['Borders:CtrBorderCount=15:CtrBorderType=Uniform:TargetBorderCount=1:TargetBorderType=MinEntropy:Prior=0/1:Prior=0.5/1:Prior=1/1', 'Counter:CtrBorderCount=15:CtrBorderType=Uniform:Prior=0/1'], 'iterations': 300, 'sampling_frequency': 'PerTree', 'fold_permutation_block': 0, 'leaf_estimation_method': 'Gradient', 'counter_calc_method': 'SkipTest', 'boosting_type': 'Plain', 'feature_border_type': 'GreedyLogSum', 'ctr_leaf_count_limit': 18446744073709551615, 'bayesian_matrix_reg': 0.1000000015, 'one_hot_max_size': 2, 'l2_leaf_reg': 9, 'random_strength': 1, 'rsm': 1, 'boost_from_average': True, 'max_ctr_complexity': 4, 'model_size_reg': 0.5, 'approx_on_full_history': False, 'simple_ctr': ['Borders:CtrBorderCount=15:CtrBorderType=Uniform:TargetBorderCount=1:TargetBorderType=MinEntropy:Prior=0/1:Prior=0.5/1:Prior=1/1', 'Counter:CtrBorderCount=15:CtrBorderType=Uniform:Prior=0/1'], 'use_best_model': True, 'random_seed': 42, 'depth': 7, 'ctr_target_border_count': 1, 'has_time': False, 'store_all_simple_ctr': False, 'fold_len_multiplier': 2, 'border_count': 254, 'bagging_temperature': 1, 'classes_count': 0, 'sparse_features_conflict_fraction': 0, 'leaf_estimation_backtracking': 'AnyImprovement', 'best_model_min_trees': 1, 'model_shrink_rate': 0, 'loss_function': 'MAE', 'learning_rate': 0.1000000015, 'score_function': 'Cosine', 'task_type': 'CPU', 'leaf_estimation_iterations': 1, 'bootstrap_type': 'Bayesian', 'permutation_count': 4}

[annaveronika]

We'll deprecate using set_param and set_params on a trained model, because some parameters are set dependant on the dataset, and it is not obvious for a user, if this parameters will be reused in a new model or reset one more time using the new dataset. So this operation will give you an error in one of the next releases.

Thanks a lot for pointing out this inconsistency!

[annaveronika]

The fix is on github now and will be out on pypi in the next release this week.

[abdullahalsaidi16]

@annaveronika Thanks a lot!