extract_sequence_features()

extract_sequence_features() builds duration, timing, and sequencing feature matrices from SequenceData. It does not run Boruta or fit a predictive model.

Function Usage

extract_sequence_features(
    seqdata,
    *,
    state_groups=None,
    timing_bin_width=12.0,
    time_unit_hint="same_as_labels",
    timing_include_start=True,
    timing_include_end=True,
    timing_count_method="any",
    timing_bin_include_left=True,
    end_time_mode="last_observed",
    sequencing_max_k=3,
    sequencing_min_support=0.05,
    sequencing_top_mined_subsequences=1000,
    sequencing_count_method="presence",
    sequencing_event_label_mode="state",
    sequencing_weighted=False,
    ids=None,
)

R / Literature Parameter Mapping

Sequenzo	R / packages	Notes
Duration block	TraMineR::seqistatd(); WeightedCluster::seqpropclust(..., properties="duration")	Spell-step totals
Timing block	Custom bins on spell start/end (Unterlerchner 2023)	No single TraMineR equivalent
Sequencing block	TraMineR::seqecreate() → TraMineR::seqefsub() → TraMineR::seqeapplysub(); WeightedCluster::seqpropclust(..., properties="pattern")	DSS spell path
Spell conversion	TraMineR::seqdss(), TraMineR::seqdur(), TraMineR::seqformat(..., to="SPELL")	Via convert_seqdata_to_spells

Entry Parameters

Parameter	Required	Type	Description
seqdata	✓	SequenceData	Input sequences.
state_groups	✗	dict / None	Map group labels to lists of states. Default: one group per state.
timing_bin_width	✗	float	Bin width in the same unit as seqdata.time (e.g. 12.0 for monthly grids, 1.0 for yearly age labels).
time_unit_hint	✗	"month" / "year" / "same_as_labels"	Metadata only; stored in results for reproducibility and self-documentation. Does not change bins; set timing_bin_width explicitly.
timing_include_start	✗	bool	Include START_* timing features.
timing_include_end	✗	bool	Include END_* timing features.
timing_count_method	✗	str	How to count events per bin (default "any").
timing_bin_include_left	✗	bool	Left-inclusive bin edges for timing.
end_time_mode	✗	"last_observed" / "exit_time"	How spell end times are defined.
sequencing_max_k	✗	int	Maximum subsequence length to mine.
sequencing_min_support	✗	int / float	Minimum support for mined subsequences.
sequencing_top_mined_subsequences	✗	int / None	Cap on number of mined subsequences (default 1000).
sequencing_count_method	✗	str	Subsequence count method (default "presence").
sequencing_event_label_mode	✗	str	Event label mode (default "state").
sequencing_weighted	✗	bool	Weighted mining; currently raises NotImplementedError.
ids	✗	list / None	Row index for output DataFrames.

Returns

A dict with:

Key	Type	Description
time_unit_hint	str	Echo of the hint argument.
timing_bin_width	float	Bin width used.
end_time_mode	str	Spell end mode used.
X_duration	DataFrame	Duration features only.
X_timing	DataFrame	Timing features only.
X_sequencing	DataFrame	Sequencing features only.
X_full	DataFrame	Horizontally stacked full matrix.
all_feature_names	list	Column names for X_full.

Example

Monthly grid (Unterlerchner-style timing)

from sequenzo import extract_sequence_features

features = extract_sequence_features(
    seqdata,
    timing_bin_width=12.0,
    time_unit_hint="month",
    timing_include_end=True,
    end_time_mode="exit_time",
    sequencing_max_k=3,
    sequencing_min_support=0.05,
)

print(features["X_full"].shape)
print(features["all_feature_names"][:5])

Yearly age labels

features = extract_sequence_features(
    seqdata,
    timing_bin_width=1.0,
    time_unit_hint="year",
    timing_include_end=True,
    end_time_mode="exit_time",
)

R Counterpart

• Closest R bundle: WeightedCluster::seqpropclust(..., prop.only=TRUE)
• Mapping note: Sequenzo implements duration, timing bins, and sequencing explicitly in Python rather than calling a single R function.

Notes

• Sequencing is mined on the spell-state sequence (DSS), not the raw panel.
• timing_bin_width=12.0 means twelve time-label units per bin, not necessarily twelve calendar months unless your grid is monthly.
• Weighted sequencing (sequencing_weighted=True) is not wired through this entrypoint.

See Also

• Section overview maps the surrounding workflow and related functions.
• Typical Workflow shows where this method fits in the full analysis.

Authors

Code: Yuqi Liang

Documentation: Yuqi Liang

References

Bolano, D., & Studer, M. (2020). The link between previous life trajectories and a later life outcome: A feature selection approach.

Unterlerchner, L., Studer, M., & Gomensoro, A. (2023). Back to the features. Investigating the relationship between educational pathways and income using sequence analysis and feature extraction and selection approach. Swiss journal of sociology, 49(2), 417-446.