Multiclassification doesn’t work. I have no ideas why. Please, help. I also tried to use target with strings instead of numbers.

target:

df['class'].value_counts()

class
0 1176
1 618
2 40
Name: count, dtype: int64

Classifier:

cb = CatBoostClassifier(loss_function='MultiClassOneVsAll',
                        random_state=RANDOM_STATE,
                        logging_level='Verbose',
                        metric_period=10,
                        name='Подбор гиперпараметров',
                        classes_count=3,
                        eval_metric='AUC:type=OneVsAll;hints=skip_train~False',
                        early_stopping_rounds=20,
                        cat_features=cat_features
                        )

param_grid:

param_grid = {'learning_rate': np.linspace(0.001, 0.40, 10),
              'iterations': [300, 500, 800, 1000, 1500, 2000],
              'max_depth': [6, 10, 16],
              'auto_class_weights': [None, 'Balanced']
              }

search

grid_search_result = cb.grid_search(param_grid,
                                    train[X],
                                    train[y],
                                    cv=3,
                                    partition_random_seed=RANDOM_STATE,
                                    calc_cv_statistics=False,
                                    search_by_train_test_split=False,
                                    refit=True,
                                    stratified=True,
                                    train_size=1,
                                    verbose=100,
                                    plot=False)

Error

CatBoostError: /src/catboost/catboost/libs/metrics/metric.cpp:3300: AUC type “OneVsAll” isn’t a singleclass AUC type

Details:

---

CatBoostError Traceback (most recent call last)
<ipython-input-99-e36ee58f1eaa> in <cell line: 4>()
2 # кроссвалидации, которую будем проводить для
3 # каждого набора гиперпараметров
4 grid_search_result = cb.grid_search(param_grid,
5 train[X],
6 train[y],

1 frames
/usr/local/lib/python3.10/dist-packages/catboost/core.py in grid_search(self, param_grid, X, y, cv, partition_random_seed, calc_cv_statistics, search_by_train_test_split, refit, shuffle, stratified, train_size, verbose, plot, plot_file, log_cout, log_cerr)
4188 raise TypeError(‘Parameter grid value is not iterable (key={!r}, value={!r})’.format(key, grid[key]))
4189
4190 return self._tune_hyperparams(
4191 param_grid=param_grid, X=X, y=y, cv=cv, n_iter=-1,
4192 partition_random_seed=partition_random_seed, calc_cv_statistics=calc_cv_statistics,

/usr/local/lib/python3.10/dist-packages/catboost/core.py in _tune_hyperparams(self, param_grid, X, y, cv, n_iter, partition_random_seed, calc_cv_statistics, search_by_train_test_split, refit, shuffle, stratified, train_size, verbose, plot, plot_file, log_cout, log_cerr)
4077
4078 with plot_wrapper(plot, plot_file, ‘Hyperparameters search plot’, [_get_train_dir(params)]):
4079 cv_result = self._object._tune_hyperparams(
4080 param_grid, train_params[“train_pool”], params, n_iter,
4081 fold_count, partition_random_seed, shuffle, stratified, train_size,

_catboost.pyx in _catboost._CatBoost._tune_hyperparams()

_catboost.pyx in _catboost._CatBoost._tune_hyperparams()

CatBoostError: /src/catboost/catboost/libs/metrics/metric.cpp:3300: AUC type “OneVsAll” isn’t a singleclass AUC type

See code fragment. I made enough amount of objects for each class