GSD TVS Evaluation

This script shows the actual performance evaluation that is run to benchmark the GSD algorithms on the TVS dataset. As this is run on the TVS dataset and has a long runtime, this example needs to be downloaded and run locally.

import os
from pathlib import Path

from joblib import Memory
from mobgap import PACKAGE_ROOT
from mobgap.data import TVSFreeLivingDataset

if "MOBGAP_TVS_DATASET_PATH" not in os.environ:
    raise ValueError(
        "Please set the environmental variable MOBGAP_TVS_DATASET_PATH to the path of the TVS dataset."
    )

dataset_path = Path(os.getenv("MOBGAP_TVS_DATASET_PATH"))
n_jobs = os.environ.get("MOBGAP_N_JOBS", 1)

free_living_data = TVSFreeLivingDataset(
    dataset_path,
    reference_system="INDIP",
    memory=Memory(PACKAGE_ROOT.parent / ".cache"),
    missing_reference_error_type="skip",
).get_subset(recording="Recording4")

Run the non-CV evaluation for all implemented algorithms

from mobgap.gait_sequences import GsdAdaptiveIonescu, GsdIluz, GsdIonescu
from mobgap.gait_sequences.pipeline import (
    GsdEmulationPipeline,
    GsdEvaluation,
    gsd_evaluation_scorer,
)

results = {}
for algo in (GsdIluz(), GsdIonescu(), GsdAdaptiveIonescu()):
    print(f"Running evaluation for {algo.__class__.__name__}")
    pipe = GsdEmulationPipeline(algo)
    eval_pipe = GsdEvaluation(
        free_living_data,
        scoring=gsd_evaluation_scorer,
        validate_paras={"n_jobs": n_jobs},
    ).run(pipe)
    results[algo.__class__.__name__] = eval_pipe.results_

Estimated memory usage: 0 MB