---

source

DataLoader

 DataLoader (max_records:int=None)

*A class for loading and processing time series data with adjustable parameters.

This class handles loading CSV data, computing returns, and calculating rolling variance.

Attributes: max_records: Maximum number of records to keep (from the end of the dataset)*

# Create a data loader with default parameters and load the data
data_loader = DataLoader(max_records=9000)
source_df = data_loader.load_data("./data/ng_daily.csv")
source_df.head()

shape: (5, 3)

date	price	ret
date	f64	f64
1997-01-07	3.82	null
1997-01-08	3.8	0.994764
1997-01-09	3.61	0.95
1997-01-10	3.92	1.085873
1997-01-13	4.0	1.020408

---

source

FeatureEngineer

 FeatureEngineer (transforms:list[__main__.DFFeature], n_shifts=3,
                  drop_nulls:bool=True)

*A class for creating lagged features from time series data.

This class handles the creation of lagged (shifted) features that can be used for GARCH-like models and other time series forecasting tasks.

Attributes: columns: List of column names to create lags for n_shifts: Number of lag periods to create drop_nulls: whether to drop the nulls after rolling window calculations*

---

source

Derivative

 Derivative (source_field:str, feature_name:str,
             requested_lag:int|None=None, step_size:int=1)

---

source

Identity

 Identity (source_field:str, feature_name:str,
           requested_lag:int|None=None, step_size:int=1)

---

source

LogReturn

 LogReturn (source_field:str, feature_name:str,
            requested_lag:int|None=None, step_size:int=1)

---

source

Square

 Square (source_field:str, feature_name:str, requested_lag:int|None=None,
         step_size:int=1)

---

source

Variance

 Variance (source_field:str, feature_name:str,
           requested_lag:int|None=None, step_size:int=1,
           rolling_variance_window:int=3)

---

source

QuantileTransformer

 QuantileTransformer (source_field:str, feature_name:str,
                      requested_lag:int|None=None, step_size:int=1,
                      n_quantiles:int=1000,
                      output_distribution:str='uniform')

A stateful quantile transformer using sklearn’s QuantileTransformer.

---

source

ZeroBasedMonth

 ZeroBasedMonth (source_field:str, feature_name:str,
                 requested_lag:int|None=None, step_size:int=1)

---

source

DFFeature

 DFFeature (source_field:str, feature_name:str,
            requested_lag:int|None=None, step_size:int=1)

feature_engineer = FeatureEngineer(
    transforms=[
        LogReturn(source_field="ret", feature_name="log_ret"),
        Variance(source_field="price", feature_name="var", requested_lag=0),
        QuantileTransformer(
            source_field="var", feature_name="var_quantile", requested_lag=0
        ),
    ],
    n_shifts=3,
)
df_with_features = feature_engineer.create_features(source_df)
df_with_features.head()

shape: (5, 9)

date	price	ret	log_ret	var	var_quantile	prev_log_ret_1	prev_log_ret_2	prev_log_ret_3
date	f64	f64	f64	f64	f64	f64	f64	f64
1997-01-13	4.0	1.020408	0.020203	0.042433	0.860861	0.082384	-0.051293	-0.005249
1997-01-14	4.01	1.0025	0.002497	0.002433	0.316111	0.020203	0.082384	-0.051293
1997-01-15	4.34	1.082294	0.079083	0.037433	0.847429	0.002497	0.020203	0.082384
1997-01-16	4.71	1.085253	0.081814	0.122633	0.944662	0.079083	0.002497	0.020203
1997-01-17	3.91	0.830149	-0.186151	0.1603	0.95773	0.081814	0.079083	0.002497

qt = QuantileTransformer(source_field="price", feature_name="price_quantile")
qt.fit(source_df)
source_df = qt.extract(source_df)
source_df.head()

shape: (5, 4)

date	price	ret	price_quantile
date	f64	f64	f64
1997-01-07	3.82	null	0.573073
1997-01-08	3.8	0.994764	0.569069
1997-01-09	3.61	0.95	0.535536
1997-01-10	3.92	1.085873	0.591091
1997-01-13	4.0	1.020408	0.607107

---

source

append_from_log_ret

 append_from_log_ret (df:polars.dataframe.frame.DataFrame,
                      new_log_ret:float, inherit_vals:list[str],
                      add_variables:dict[str,float])

*Adds a new record to the dataframe based on a log return value.

Args: df: Input DataFrame containing time series data new_log_ret: The new log return value to add

Returns: DataFrame with a new row appended*

---

source

binary_feature_from_date_ranges

 binary_feature_from_date_ranges
                                  (date_range:tuple[datetime.date,datetime
                                  .date], periods:list[tuple[datetime.date
                                  ,datetime.date]],
                                  feature_name:str='feature')

binary_feature_from_date_ranges(
    date_range=(date(2010, 1, 1), date(2026, 1, 1)),
    periods=[
        (date(2022, 2, 24), date(2026, 1, 1)),
    ],
    feature_name="RU/UA_war",
)

shape: (5_845, 2)

date	RU/UA_war
date	i64
2010-01-01	0
2010-01-02	0
2010-01-03	0
2010-01-04	0
2010-01-05	0
…	…
2025-12-28	1
2025-12-29	1
2025-12-30	1
2025-12-31	1
2026-01-01	1