AttributeError: 'Pipeline' object has no attribute 'features_in'. Did you mean: 'feature_list'?

[LuisAlejandro]

I'm having this error using the fretrade/freqai bot.

2025-08-22 07:56:12,576 - freqtrade - ERROR - Fatal exception!
Traceback (most recent call last):
  File "/freqtrade/freqtrade/main.py", line 47, in main
    return_code = args["func"](args)
  File "/freqtrade/freqtrade/commands/optimize_commands.py", line 61, in
start_backtesting
    backtesting.start()
    ~~~~~~~~~~~~~~~~~^^
  File "/freqtrade/freqtrade/optimize/backtesting.py", line 1820, in start
    min_date, max_date = self.backtest_one_strategy(strat, data, timerange)
                         ~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
  File "/freqtrade/freqtrade/optimize/backtesting.py", line 1730, in
backtest_one_strategy
    preprocessed = self.strategy.advise_all_indicators(data)
  File "/freqtrade/freqtrade/strategy/interface.py", line 1748, in
advise_all_indicators
    res[pair] = self.advise_indicators(pair_data.copy(), {"pair": pair}).copy()
                ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/freqtrade/freqtrade/strategy/interface.py", line 1803, in
advise_indicators
    dataframe = self.populate_indicators(dataframe, metadata)
  File "/freqtrade/user_data/strategies/EnhancedLSTMStrategy.py", line 441, in
populate_indicators
    dataframe = self.freqai.start(dataframe, metadata, self)
  File "/freqtrade/freqtrade/freqai/freqai_interface.py", line 161, in start
    dk = self.start_backtesting(dataframe, metadata, self.dk, strategy)
  File "/freqtrade/freqtrade/freqai/freqai_interface.py", line 392, in
start_backtesting
    pred_df, do_preds = self.predict(dataframe_backtest, dk)
                        ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
  File "/freqtrade/freqtrade/freqai/base_models/BasePyTorchRegressor.py", line
43, in predict
    dk.data_dictionary["prediction_features"], outliers, _ =
dk.feature_pipeline.transform(
                                                             ~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~^
        dk.data_dictionary["prediction_features"], outlier_check=True
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File
"/home/freqtrade/.local/lib/python3.13/site-packages/datasieve/pipeline.py",
line 92, in transform
    X, y, sample_weight = self._validate_arguments(
                          ~~~~~~~~~~~~~~~~~~~~~~~~^
        X, y, sample_weight, outlier_check=outlier_check)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File
"/home/freqtrade/.local/lib/python3.13/site-packages/datasieve/pipeline.py",
line 167, in _validate_arguments
    if list(X.columns) != list(self.features_in):
                               ^^^^^^^^^^^^^^^^
AttributeError: 'Pipeline' object has no attribute 'features_in'. Did you mean:
'feature_list'?

I'm still figuring out what is happening, but regardless of that, I think there might me an error on the Pipeline class. Apparently it references features_in but its value is never initialized, or thats what I saw, maybe I'm wrong.

Greetings.

[robcaulk]

Hello,

Please show how you are using DataSieve objects. We cannot be responsible for personal code embedded in other projects and how they call/use objects.

features_in is defined during the argument validation: https://github.com/emergentmethods/datasieve/blob/main/datasieve/pipeline.py#L151

[LuisAlejandro]

but what happens if I call _validate_arguments with X=DataFrame and fit=False?

it goes to this block where features_in hasn't been defined yet but is referenced:

        elif isinstance(X, pd.DataFrame) and not fit:
            if list(X.columns) != list(self.features_in):
                raise Exception(f"Pipeline expected {self.features_in} but got {X.columns}.")

don't worry about my bug, I opened the ticket to explore a potential bug in DataSieve.

you can close it if you don't see it worth exploring/pursuing.

[LuisAlejandro]

I believe that use case (X=DataFrame, fit=False) happens right here:

https://github.com/emergentmethods/datasieve/blob/main/datasieve/pipeline.py#L92

If I have some spare time this next week I can contribute some tests and see whats going on

[robcaulk]

but what happens if I call _validate_arguments with X=DataFrame and fit=False?

I’m not sure why you would do this before fitting the object? If you haven’t fit the pipeline, then the pipeline won’t have any transformations to make on your data.

You need to fit the pipeline before you can call it as a transform. This isn’t unique to data sieve - this is just standard sklearn/data pipelining.

Let me know if I’m missing something?

[SimonAltmannshofer]

I have the same problem when I run the freqai example in backtesting mode:
freqtrade backtesting --config config_examples/config_freqai.example.json --strategy FreqaiExampleStrategy --freqaimodel PyTorchMLPRegressor --strategy-path freqtrade/templates --timerange 20250101-20250130

Result:
2026-02-27 23:05:59,155 - freqtrade - ERROR - Fatal exception!
Traceback (most recent call last):
File "/home/simon/freqtrade/freqtrade/main.py", line 53, in main
return_code = args"func"
^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/freqtrade/commands/optimize_commands.py", line 61, in start_backtesting
backtesting.start()
File "/home/simon/freqtrade/freqtrade/optimize/backtesting.py", line 1843, in start
min_date, max_date = self.backtest_one_strategy(strat, data, timerange)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/freqtrade/optimize/backtesting.py", line 1753, in backtest_one_strategy
preprocessed = self.strategy.advise_all_indicators(data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/freqtrade/strategy/interface.py", line 1749, in advise_all_indicators
res[pair] = self.advise_indicators(pair_data.copy(), {"pair": pair}).copy()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/freqtrade/strategy/interface.py", line 1811, in advise_indicators
dataframe = self.populate_indicators(dataframe, metadata)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/freqtrade/templates/FreqaiExampleStrategy.py", line 233, in populate_indicators
dataframe = self.freqai.start(dataframe, metadata, self)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/freqtrade/freqai/freqai_interface.py", line 161, in start
dk = self.start_backtesting(dataframe, metadata, self.dk, strategy)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/freqtrade/freqai/freqai_interface.py", line 392, in start_backtesting
pred_df, do_preds = self.predict(dataframe_backtest, dk)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/freqtrade/freqai/base_models/BasePyTorchRegressor.py", line 43, in predict
dk.data_dictionary["prediction_features"], outliers, _ = dk.feature_pipeline.transform(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/.venv/lib/python3.12/site-packages/datasieve/pipeline.py", line 92, in transform
X, y, sample_weight = self._validate_arguments(
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/simon/freqtrade/.venv/lib/python3.12/site-packages/datasieve/pipeline.py", line 167, in _validate_arguments
if list(X.columns) != list(self.features_in):
^^^^^^^^^^^^^^^^
AttributeError: 'Pipeline' object has no attribute 'features_in'. Did you mean: 'feature_list'?

When I change the timerange, it works:
freqtrade backtesting --config config_examples/config_freqai.example.json --strategy FreqaiExampleStrategy --freqaimodel PyTorchMLPRegressor --strategy-path freqtrade/templates --timerange 20251010-20251020

I think I downloaded enough data with this:
freqtrade download-data --config config_examples/config_freqai.example.json --timerange 20240101-20251231

[luckykefu]

On my computer, I downloaded more OHLCV data, which resolved the error. I believe the training range was too large, exceeding the current historical data. Therefore, you could try reducing the training range: train_period_days + startup_candle_count × max_timeframe. For example, 60 days + 80 × 3 days = 300. The training range would be approximately 20220101, pushing back 300 days to 20210306. Your local data should start from 20210306. However, this is likely a bug or a logical flaw.