Catboost: Why is multiclass classification internally transforming to regression/single class classification problem

Question

While using a custom loss function in a multi class classification, I am getting an error that my custom objective function is not having calc_ders_range attribute. However, as per my discussion in Telegram channel of catboost, calc_ders_range is for single classification/regression. I am puzzled to get the error below even when I am passing my custom objective to CatBoostClassifier.

My Code: Output label is of int64 type with values from 0 to 25 representing 26 classes. Custom objective and accuracy metric are taken from Usage Examples for sample https://catboost.ai/docs/concepts/python-usages-examples.html#user-defined-loss-function

class MyObjective(object):
    def calc_ders_multi(self, approx, target, weight):
        approx = np.array(approx) - max(approx)
        exp_approx = np.exp(approx)
        exp_sum = exp_approx.sum()
        grad = []
        hess = []
        for j in range(len(approx)):
            der1 = -exp_approx[j] / exp_sum
            if j == target:
                der1 += 1
            hess_row = []
            for j2 in range(len(approx)):
                der2 = exp_approx[j] * exp_approx[j2] / (exp_sum**2)
                if j2 == j:
                    der2 -= exp_approx[j] / exp_sum
                hess_row.append(der2 * weight)
                
            grad.append(der1 * weight)
            hess.append(hess_row)
            
        return (grad, hess)

class AccuracyMetric(object):
    def get_final_error(self, error, weight):
        return error / (weight + 1e-38)

    def is_max_optimal(self):
        return True

    def evaluate(self, approxes, target, weight):
        best_class = np.argmax(approxes, axis=0)
        
        accuracy_sum = 0
        weight_sum = 0 

        for i in range(len(target)):
            w = 1.0 if weight is None else weight[i]
            weight_sum += w
            accuracy_sum += w * (best_class[i] == target[i])

        return accuracy_sum, weight_sum

def get_pipeline(args):
    """Create a pipeline."""
    pipeline_feat1 = Pipeline([
        ('selector', ColumnSelector(cols='feat1', drop_axis=True)),
        ('vec', TfidfVectorizer(tokenizer=word_tokenize)),
    ])
    pipeline_feat2 = Pipeline([
        ('selector', ColumnSelector(cols='feat2', drop_axis=False)),
        ('imputer', SimpleImputer(missing_values=np.nan, strategy="constant", fill_value=0, copy=False)),
        ('ohe', OneHotEncoder(handle_unknown='ignore')),
    ])
    pipeline_feat3 = Pipeline([
        ('selector', ColumnSelector(cols='feat3', drop_axis=False)),
        ('imputer', SimpleImputer(missing_values=np.nan, strategy="constant", fill_value=0, copy=False)),
        ('ohe', OneHotEncoder(handle_unknown='ignore')),
    ])
    features = FeatureUnion([
        ('f1', pipeline_feat1),
        ('f2', pipeline_feat2),
        ('f3', pipeline_feat3),
    ])

    steps = [
        ('features', features),
        ('clf', CatBoostClassifier(task_type='CPU', iterations=5000, random_seed=0, 
loss_function=MyObjective(), eval_metric=AccuracyMetric(), verbose=100))
    ]
    train_pipeline = Pipeline(steps)

    params = {
        "features__f1__vec__max_features": args.f1_max_features,
        "features__f1__vec__ngram_range": (1, args.f1_max_ngram)
    }
    params = {k: v for k, v in params.items() if v is not None}

    train_pipeline.set_params(**params)
    return train_pipeline

# Train model.
pipeline = get_pipeline()
# Split train and test data.
X_train, X_val, y_train, y_val = train_test_split(df_train[['feat1', 'feat2', 'feat3']], df_train['label'], train_size=0.8, random_state=21)
model = pipeline.fit(X_train, y_train)

Error Message:

AttributeError                            Traceback (most recent call last)
_catboost.pyx in _catboost._ObjectiveCalcDersRange()

AttributeError: 'MyObjective' object has no attribute 'calc_ders_range'

During handling of the above exception, another exception occurred:

CatBoostError                             Traceback (most recent call last)
<ipython-input-12-ea20f154d788> in <module>
     10                                                     train_size=0.8,
     11                                                     random_state=21)
---> 12 model = pipeline.fit(X_train, y_train)
     13 

/opt/conda/lib/python3.7/site-packages/sklearn/pipeline.py in fit(self, X, y, **fit_params)
    333             if self._final_estimator != 'passthrough':
    334                 fit_params_last_step = fit_params_steps[self.steps[-1][0]]
--> 335                 self._final_estimator.fit(Xt, y, **fit_params_last_step)
    336 
    337         return self

/opt/conda/lib/python3.7/site-packages/catboost/core.py in fit(self, X, y, cat_features, text_features, embedding_features, sample_weight, baseline, use_best_model, eval_set, verbose, logging_level, plot, column_description, verbose_eval, metric_period, silent, early_stopping_rounds, save_snapshot, snapshot_file, snapshot_interval, init_model)
   4296         self._fit(X, y, cat_features, text_features, embedding_features, None, sample_weight, None, None, None, None, baseline, use_best_model,
   4297                   eval_set, verbose, logging_level, plot, column_description, verbose_eval, metric_period,
-> 4298                   silent, early_stopping_rounds, save_snapshot, snapshot_file, snapshot_interval, init_model)
   4299         return self
   4300 

/opt/conda/lib/python3.7/site-packages/catboost/core.py in _fit(self, X, y, cat_features, text_features, embedding_features, pairs, sample_weight, group_id, group_weight, subgroup_id, pairs_weight, baseline, use_best_model, eval_set, verbose, logging_level, plot, column_description, verbose_eval, metric_period, silent, early_stopping_rounds, save_snapshot, snapshot_file, snapshot_interval, init_model)
   1807                 params,
   1808                 allow_clear_pool,
-> 1809                 train_params["init_model"]
   1810             )
   1811 

/opt/conda/lib/python3.7/site-packages/catboost/core.py in _train(self, train_pool, test_pool, params, allow_clear_pool, init_model)
   1256 
   1257     def _train(self, train_pool, test_pool, params, allow_clear_pool, init_model):
-> 1258         self._object._train(train_pool, test_pool, params, allow_clear_pool, init_model._object if init_model else None)
   1259         self._set_trained_model_attributes()
   1260 

_catboost.pyx in _catboost._CatBoost._train()

_catboost.pyx in _catboost._CatBoost._train()

CatBoostError: catboost/python-package/catboost/helpers.cpp:42: Traceback (most recent call last):
  File "_catboost.pyx", line 1345, in _catboost._ObjectiveCalcDersRange
AttributeError: 'MyObjective' object has no attribute 'calc_ders_range'

Answer 1

I managed to solve this problem by explicitly setting the classes_count parameter. e.g.

CatBoostClassifier(
        iterations=10,
        depth=2,
        classes_count=17