SensorPositionComparison - Full Pipeline#

Algorithms#

Hide code cell source 
from myst_nb_bokeh import glue_bokeh

from gaitmap_bench import config, is_config_set
from gaitmap_bench.docu_utils import set_docs_config
from gaitmap_challenges.full_pipeline.sensor_position_comparison_instep import Challenge
from gaitmap_challenges.results import load_run, get_latest_result, filter_results, get_all_result_paths, \
    generate_overview_table
from gaitmap_challenges.visualization import SingleMetricBoxplot, group_by_data_label

is_config_set() or set_docs_config()
all_runs = get_all_result_paths(Challenge, config().results_dir)
all_runs = filter_results(all_runs, challenge_version=Challenge.VERSION, is_debug_run=False)
latest_runs = get_latest_result(all_runs)
generate_overview_table(latest_runs).set_index("Entry").T
Entry (gaitmap, mad_classic, default) (gaitmap, mad_modern, default) (gaitmap, mad_modern, optimized_zupt)
Datetime 2023-07-18T14:45:11.258248+02:00 2023-07-18T14:45:12.239749+02:00 2023-07-18T14:45:12.241794+02:00
Description Default MaD pipeline Modern MaD pipeline Modern MaD pipeline with optimized ZUPT detection
References [https://ieeexplore.ieee.org/document/6949634,... [https://doi.org/10.1186/s12984-021-00883-7, h... [https://doi.org/10.1186/s12984-021-00883-7, h...
Code Authors [MaD-DiGait] [MaD-DiGait] [MaD-DiGait]
Algorithm Authors [See source and references for individual algo... [See source and references for individual algo... [See source and references for individual algo...
Implementation https://github.com/mad-lab-fau/gaitmap https://github.com/mad-lab-fau/gaitmap https://github.com/mad-lab-fau/gaitmap

Absolute Errors of the Mean#

The main outcome are the absolute errors of the mean gait parameters. This means one error value per gait test is calculated. The first plot pools all datapoints across all 5 test folds.

The second plot presents the mean absolute error over each fold.

Results per Participant and Test#

Hide code cell source 
from gaitmap_bench.docu_utils import glue_bokeh_md, tabs
from myst_nb import glue
from IPython.display import Markdown

run_info = {k: load_run(Challenge, v) for k, v in latest_runs.items()}
cv_results = {k: v.results["cv_results"]  for k, v in run_info.items()}

tab_items = {}
metrics = {
    "Gait Velocity": "gait_velocity",
    "Stride Length": "stride_length",
    "Stride Time": "stride_time",
    "Swing Time": "swing_time",
}

for name, metric in metrics.items():
    p = SingleMetricBoxplot(cv_results, f"{metric}__abs_error", "single", overlay_scatter=True, label_grouper=group_by_data_label(level="test", include_all="Combined"))
    glue_name = f"single_{metric}"
    glue_bokeh(glue_name, p.bokeh())
    tab_items[name] = glue_bokeh_md(glue_name)

glue("single_results", Markdown(tabs(tab_items, class_str="full-width")), display=False)

Results per Participant and Test (inverted)#

Hide code cell source 
for name, metric in metrics.items():
    p = SingleMetricBoxplot(cv_results, f"{metric}__abs_error", "single", overlay_scatter=True, label_grouper=group_by_data_label(level="test", include_all="Combined"), invert_grouping=True)
    glue_name = f"single_inverted_{metric}"
    glue_bokeh(glue_name, p.bokeh())
    tab_items[name] = glue_bokeh_md(glue_name)

glue("single_inverted_results", Markdown(tabs(tab_items, class_str="full-width")), display=False)

Results per CV Fold#

Hide code cell source 
tab_items = {}

for name, metric in metrics.items():
    p = SingleMetricBoxplot(cv_results, f"agg__{metric}__abs_error_mean", "fold", overlay_scatter=True)
    glue_name = f"fold_{metric}"
    glue_bokeh(glue_name, p.bokeh())
    tab_items[name] = glue_bokeh_md(glue_name)

glue("fold_results", Markdown(tabs(tab_items, class_str="full-width")), display=False)

Residual Plots#

Below residual plots for each algorithm and metric are shown to further invesqtigate the dependency of the error on the actual measurement value.

Hide code cell source 
from gaitmap_challenges.visualization import ResidualPlot

final_markdown = []

for algo, results in cv_results.items():
    tab_items = {}
    for name, metric in metrics.items():
        p = ResidualPlot(results, prediction_col_name=f"{metric}__predicted", reference_col_name=f"{metric}__reference", metric_name=name)
        glue_name = f"residual_{algo}_{metric}"
        glue_bokeh(glue_name, p.bokeh())
        tab_items[name] = glue_bokeh_md(glue_name)
    final_markdown.append(f"#### {algo}")
    final_markdown.append(tabs(tab_items, class_str="full-width", sync=True))

glue("all_residual", Markdown("\n".join(final_markdown)), display=False)

(‘gaitmap’, ‘mad_classic’, ‘default’)#

(‘gaitmap’, ‘mad_modern’, ‘default’)#

(‘gaitmap’, ‘mad_modern’, ‘optimized_zupt’)#