StandardSequencer(timestamp_column: str, category_column: str = None, overwrite: Dict[str, str] = None, resampling_freq: str = '1D', gap_threshold: int = 60000, sequence_length: int = 4, sequence_shift: int = 1, **kwargs)
: Base class for all sequencer steps. These steps are used to
specify transformation and filling operations on timeseries datasets
that occur before the data preprocessing takes place.
Initializing the StandardSequencer step, which is responsible for extracting sequences from any timeseries dataset. The main logic behind this step can be summed up in a few steps as follows: 1. First, we define how to add the corresponding timestamp to each datapoint by using the function `get_timestamp_do_fn`. In this implementation, the timestamp is expected to be a unix timestamp. 2. Similarly, we define how to add a categorical key to each datapoint if a categorical column is provided. 3. Following that, we use the timestamp, the categorical key and the gap threshold to split the data into so-called 'sessions' 4. Once the data is split into sessions, we resample the sessions based on the `resampling_freq` to create equidistant timestamps, fill the missing values and extract the finalized sequences based on the `sequence_length` and `sequence_shift` :param timestamp_column: string, the name of the column for the timestamp resides :param category_column: string, the name of the column of a possible categorical feature :param overwrite: dict, used to overwrite any of the default resampling and filling behaviour :param resampling_freq: string, the resampling frequency as an Offset Alias :param gap_threshold: int, the minimum gap between two sessions in seconds :param sequence_length: int, the desired length of a sequence in terms of datapoints :param sequence_shift: int, the number steps to shift before extracting the next sequence :param kwargs: additional params ### Ancestors (in MRO) * zenml.core.steps.sequencer.base_sequencer.BaseSequencerStep * zenml.core.steps.base_step.BaseStep ### Methods `get_category_do_fn(self)` : Creates a class which inherits from beam.DoFn to add a categorical key to each datapoint :return: an instance of the beam.DoFn `get_combine_fn(self)` : Creates a class which inherits from beam.CombineFn which processes sessions and extracts sequences from it :return: an instance of the beam.CombineFn `get_timestamp_do_fn(self)` : Creates a class which inherits from beam.DoFn to add the timestamp to each datapoint :return: an instance of the beam.DoFn `get_window(self)` : Returns a selected beam windowing strategy :return: the selected windowing strategy