---
metadata:
  - name: generator
    content: Diplodoc Platform v5.54.5
alternate:
  - https://catboost.ai/docs/en/concepts/python-reference_sum_models.md
  - href: en/concepts/python-reference_sum_models.md
    type: text/markdown
    title: Markdown version
  - href: ../llms.txt
    type: text/markdown
    title: llms.txt
---
> **Documentation Index:** Fetch the complete configuration index at https://catboost.ai/docs/en/llms.txt

# sum_models

## Purpose {#purpose}

<!-- source: en/_includes/work_src/reusage-python/python__sum-limits__desc.md -->
Blend trees and counters of two or more trained CatBoost models into a new model. Leaf values can be individually weighted for each input model. For example, it may be useful to blend models trained on different validation datasets.
<!-- endsource: en/_includes/work_src/reusage-python/python__sum-limits__desc.md -->


## Method call format {#call-format}

```python
sum_models(models,
           weights=None,
           ctr_merge_policy='IntersectingCountersAverage')
```

## Parameters {#parameters}

<!-- source: en/_includes/work_src/reusage-python/python__sum-limits__parameters.md -->
 ### models

#### Description

A list of models to blend.

**Possible values**

list of CatBoost models

**Default value**

Required parameter

### weights

#### Description

A list of weights for the leaf values of each model. The length of this list must be equal to the number of blended models.

А list of weights equal to <q>1.0/N</q> for N blended models gives the average prediction. For example, the following list of weights gives the average prediction for four blended models:

```
[0.25,0.25,0.25,0.25]
```
**Possible values**

list of numbers

**Default value**

None (leaf values weights are set to 1 for all models)

### ctr_merge_policy

#### Description

The counters merging policy. Possible values:
- FailIfCtrIntersects — Ensure that the models have zero intersecting counters.
- LeaveMostDiversifiedTable — Use the most diversified counters by the count of unique hash values.
- IntersectingCountersAverage — Use the average ctr counter values in the intersecting bins.
- KeepAllTables — Keep Counter and FeatureFreq ctr's from all models.

**Possible values**

string

**Default value**

IntersectingCountersAverage
<!-- endsource: en/_includes/work_src/reusage-python/python__sum-limits__parameters.md -->


{% note info %}

- The bias of the models sum is equal to the weighted sum of models biases.
- The scale of the models sum is equal to 1, leaf values are scaled before the summation.

{% endnote %}

## Type of return value {#usage-example}

CatBoost model

## Example {#example}

```python
from catboost import CatBoostClassifier, Pool, sum_models
from catboost.datasets import amazon
import numpy as np
from sklearn.model_selection import train_test_split

train_df, _ = amazon()

y = train_df.ACTION
X = train_df.drop('ACTION', axis=1)

categorical_features_indices = np.where(X.dtypes != np.float)[0]

X_train, X_validation, y_train, y_validation = train_test_split(X,
                                                                y,
                                                                train_size=0.8,
                                                                random_state=42)

train_pool = Pool(X_train,
                  y_train,
                  cat_features=categorical_features_indices)
validate_pool = Pool(X_validation,
                     y_validation,
                     cat_features=categorical_features_indices)

models = []
for i in range(5):
    model = CatBoostClassifier(iterations=100,
                               random_seed=i)
    model.fit(train_pool,
              eval_set=validate_pool)
    models.append(model)

models_avrg = sum_models(models,
                         weights=[1.0/len(models)] * len(models))
```