[Bug]: cannot get prediction for boosting models

[DRMPN]

Expected Behavior

Prediction can be obtained after pipeline fitting for boosting models.

Current Behavior

Cannot get prediction for regression boosting models after fitting.

For xgboostreg:

2024-09-01 13:12:59,554 - PipelineNode - Obtain prediction in pipeline node by operation: xgboostreg
Traceback (most recent call last):
  File "c:\Users\nnikitin-user\Desktop\automl-september\run_xgboost\run_xgboost.py", line 36, in <module>
    prediction = auto_model.predict(features=test)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\api\main.py", line 297, in predict
    self.prediction = self.data_processor.define_predictions(current_pipeline=self.current_pipeline,
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\api\api_utils\api_data.py", line 102,
in define_predictions
    output_prediction = current_pipeline.predict(test_data)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\pipelines\pipeline.py", line 285, in predict
    result = self.root_node.predict(input_data=copied_input_data, output_mode=output_mode)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\pipelines\node.py", line 231, in
predict
    operation_predict = self.operation.predict(fitted_operation=self.fitted_operation,
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\operation.py", line 105, in predict
    return self._predict(fitted_operation, data, params, output_mode, is_fit_stage=False)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\operation.py", line 133, in _predict
    prediction = self._eval_strategy.predict(
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\evaluation\boostings.py", line 73, in predict
    prediction = trained_operation.predict(predict_data)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\evaluation\operation_implementations\models\boostings_implementations.py", line 65, in predict
    X, _ = self.convert_to_dataframe(input_data, self.params.get('enable_categorical'))
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\evaluation\operation_implementations\models\boostings_implementations.py", line 95, in convert_to_dataframe
    dataframe['target'] = np.ravel(data.target)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\frame.py", line 3980, in __setitem__
    self._set_item(key, value)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\frame.py", line 4174, in _set_item
    value = self._sanitize_column(value)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\frame.py", line 4915, in _sanitize_column
    com.require_length_match(value, self.index)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\common.py", line 571, in require_length_match
    raise ValueError(
ValueError: Length of values (0) does not match length of index (125690)

For lgbmreg:

2024-09-01 20:54:37,325 - PipelineNode - Obtain prediction in pipeline node by operation: lgbmreg
Traceback (most recent call last):
  File "c:\Users\nnikitin-user\Desktop\automl-september\run_lgbm\run_lgbm.py", line 39, in <module>
    prediction = auto_model.predict(features=test)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\api\main.py", line 297, in predict
    self.prediction = self.data_processor.define_predictions(current_pipeline=self.current_pipeline,
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\api\api_utils\api_data.py", line 102,
in define_predictions
    output_prediction = current_pipeline.predict(test_data)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\pipelines\pipeline.py", line 285, in predict
    result = self.root_node.predict(input_data=copied_input_data, output_mode=output_mode)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\pipelines\node.py", line 231, in
predict
    operation_predict = self.operation.predict(fitted_operation=self.fitted_operation,
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\operation.py", line 105, in predict
    return self._predict(fitted_operation, data, params, output_mode, is_fit_stage=False)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\operation.py", line 133, in _predict
    prediction = self._eval_strategy.predict(
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\evaluation\boostings.py", line 73, in predict
    prediction = trained_operation.predict(predict_data)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\evaluation\operation_implementations\models\boostings_implementations.py", line 202, in predict
    X, _ = self.convert_to_dataframe(input_data, identify_cats=self.params.get('enable_categorical'))
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\fedot\core\operations\evaluation\operation_implementations\models\boostings_implementations.py", line 240, in convert_to_dataframe
    dataframe['target'] = np.ravel(data.target)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\frame.py", line 3980, in __setitem__
    self._set_item(key, value)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\frame.py", line 4174, in _set_item
    value = self._sanitize_column(value)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\frame.py", line 4915, in _sanitize_column
    com.require_length_match(value, self.index)
  File "C:\Users\nnikitin-user\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\common.py", line 571, in require_length_match
    raise ValueError(
ValueError: Length of values (0) does not match length of index (125690)

Possible Solution

At first glance, I thought the hardcoded "target" was a problem. However, when I tested it, the name of the target column didn't matter at all.

Steps to Reproduce

Data can be obtained from: https://www.kaggle.com/competitions/playground-series-s4e9/data
Simple notebook to reproduce the bug:

import pandas as pd
import numpy as np
from fedot.api.main import Fedot
from fedot.core.pipelines.pipeline_builder import PipelineBuilder

train = pd.read_csv("C:/Users/nnikitin-user/Desktop/automl-september/playground-series-s4e9/train.csv")
test = pd.read_csv("C:/Users/nnikitin-user/Desktop/automl-september/playground-series-s4e9/test.csv")
sub = pd.read_csv("C:/Users/nnikitin-user/Desktop/automl-september/playground-series-s4e9/sample_submission.csv")

train.drop(columns=["id"], inplace=True)
test.drop(columns=["id"], inplace=True)

auto_model = Fedot(
    problem="regression",
    metric=["rmse"],
    preset="best_quality",
    with_tuning=True,
    timeout=5,
    cv_folds=10,
    seed=42,
    n_jobs=1,
    logging_level=10,
    initial_assumption=PipelineBuilder().add_node("xgboostreg").build(),
    use_pipelines_cache=False,
    use_auto_preprocessing=False,
)

auto_model.fit(features=train, target="price")

auto_model.current_pipeline.save(
    path="C:/Users/nnikitin-user/Desktop/automl-september/run_xgboost/saved_pipelines",
    create_subdir=True,
    is_datetime_in_path=True,
)

prediction = auto_model.predict(features=test)

sub["price"] = prediction.ravel()
sub.to_csv("submission.csv", index=False)

Context [OPTIONAL]

Participating in a Kaggle competition https://www.kaggle.com/competitions/playground-series-s4e9

[aPovidlo]

@DRMPN Да, действительно это баг. При классификации не падает в convert_to_dataset видимо из-за того, что во время predict data.target = None. А при регрессии он чем-то заполнен, скорее всего np.array([]), поэтому и падает при попытке записать данные в dataframe для бустинговых моделей. Fix я думаю тут в том, чтобы добавить новое условие сюда:
if data.target is not None or data.target.size != 0:

[DRMPN]

@DRMPN Да, действительно это баг. При классификации не падает в convert_to_dataset видимо из-за того, что во время predict data.target = None. А при регрессии он чем-то заполнен, скорее всего np.array([]), поэтому и падает при попытке записать данные в dataframe для бустинговых моделей. Fix я думаю тут в том, чтобы добавить новое условие сюда: if data.target is not None or data.target.size != 0:

Не помогло, проблема где-то еще.

[aPovidlo]

А ошибка все таже?

[DRMPN]

А ошибка все таже?

Да

[aPovidlo]

У меня получилось если добавить условие с data.target.size > 0. Добавил его не только для XGBoost, но и для LightGBM. Там тоже потенциально может возникнуть такая ошибка

[aPovidlo]

@DRMPN думаю, что я починил эту проблему в #1320. Запустил код воспроизведения и он отработал штатно в том PR.

[DRMPN]

Отлично, тогда я еще раз проверю и закрою, когда ПР будет влит в мастер!

[DRMPN]

Tested, the bug is fixed in #1320.
Attached log files for the test run: logs.txt