sport-activities-features 0.2.12

A minimalistic toolbox for extracting features from sport activity files

---

sport-activities-features --- A minimalistic toolbox for extracting features from sport activity files written in Python

---

General outline of the framework

Monitoring of sport activities produces many geographic, topologic and personalized data, with a vast majority of details hidden. Thus, a rigorous ex-post data analysis and statistic evaluation are required to extract them. Namely, most of the mainstream solutions for analyzing sport activities files rely on integral metrics, such as total duration, total distance and average hearth rate, which may suffer of "overall (integral) metrics problem". Among others, such problems are expressed in capturing sport activities in general only (ommiting crucial components), calculating potentially fallacious and misleading metrics, not recognizing different stages/phases of the sport activity (warm-up, endurance, intervals), and others.

The sport-activities-framework on the other side offers a detailed insight into the sport activity files. The framework supports both identification and extraction methods, such as identifying number of hills, extracting the average altitudes of identified hills, measuring total distance of identified hills, deriving climbing ratios (total distance of identified hills vs. total distance), average/total ascents of hills and so much more. The framework also integrates many other extensions, among others historical weather parsing, statistical evaluations and ex-post visualizations. Previous work on these topical questions were addressed in relevant scientific papers on data mining, also in a combination with the generating/predicting automated sport training sessions.

Detailed insights

The sport-activities-features framework is compatible with TCX & GPX activity files and Overpass API nodes. Current version witholds (but is not limited to) following functions:

• extracting integral metrics, such as total distance, total duration, calories (see example),
• extracting topographic features, such as number of hills, average altitude of identified hills, total distance of identified hills, climbing ratio, average ascent of hills, total ascent, total descent (see example),
• plotting identified hills (see example),
• extracting the intervals, such as number of intervals, maximum/minimum/average duration of intervals, maximum/minimum/average distance of intervals, maximum/minimum/average heart rate during intervals,
• plotting the identified intervals (see example),
• calculating the training loads, such as Bannister TRIMP, Lucia TRIMP(see example),
• parsing the historical weather data from an external service,
• extracting the integral metrics of the activity inside area given with coordinates (distance, heartrate, speed) (see example),
• extracting the activities from CSV file(s) and randomly selecting the specific number of activities (see example),
• extracting the dead ends,
• and much more.

The framework comes with two (testing) benchmark datasets, which are freely available to download from: DATASET1, DATASET2.

Installation

pip3

Install sport-activities-features with pip3:

pip3 install sport-activities-features

Fedora Linux

To install sport-activities-features on Fedora, use:

$ dnf install python3-sport-activities-features

API

There is a simple API for remote work with sport-activities-features package available here.

Historical weather data

Weather data parsed is collected from the Visual Crossing Weather API. Please note that this is an external unaffiliated service and users must register to use the API. The service has a free tier (1000 Weather reports / day) but is otherwise operating on a pay-as-you-go model. For pricing and terms of use please read the official documentation of the API provider.

Overpass API & Open Elevation API integration

Without performed activities we can use the OpenStreetMap for identification of hills, total ascent and descent. This is done using the Overpass API which is a read-only API that allows querying of OSM map data. In addition to that altitude data is retrieved by using the Open-Elevation API which is a open-source and free alternative to the Google Elevation API. Both of the solutions can be used by using free publicly acessible APIs (Overpass, Open-Elevation) or can be self hosted on a server or as a Docker container (Overpass, Open-Elevation).

CODE EXAMPLES:

Reading files

(*.TCX)

from sport_activities_features.tcx_manipulation import TCXFile

# Class for reading TCX files
tcx_file=TCXFile()
data = tcx_file.read_one_file("path_to_the_file")

(*.GPX)

from sport_activities_features.gpx_manipulation import GPXFile

# Class for reading GPX files
gpx_file=GPXFile()

# Read the file and generate a dictionary with 
data = gpx_file.read_one_file("path_to_the_file")

Extraction of topographic features

from sport_activities_features.hill_identification import HillIdentification
from sport_activities_features.tcx_manipulation import TCXFile
from sport_activities_features.topographic_features import TopographicFeatures
from sport_activities_features.plot_data import PlotData

# Read TCX file
tcx_file = TCXFile()
activity = tcx_file.read_one_file("path_to_the_file")

# Detect hills in data
Hill = HillIdentification(activity['altitudes'], 30)
Hill.identify_hills()
all_hills = Hill.return_hills()

# Extract features from data
Top = TopographicFeatures(all_hills)
num_hills = Top.num_of_hills()
avg_altitude = Top.avg_altitude_of_hills(activity['altitudes'])
avg_ascent = Top.avg_ascent_of_hills(activity['altitudes'])
distance_hills = Top.distance_of_hills(activity['positions'])
hills_share = Top.share_of_hills(distance_hills, activity['total_distance'])

Extraction of intervals

import sys
sys.path.append('../')

from sport_activities_features.interval_identification import IntervalIdentificationByPower, IntervalIdentificationByHeartrate
from sport_activities_features.tcx_manipulation import TCXFile

# Reading the TCX file
tcx_file = TCXFile()
activity = tcx_file.read_one_file("path_to_the_data")

# Identifying the intervals in the activity by power
Intervals = IntervalIdentificationByPower(activity["distances"], activity["timestamps"], activity["altitudes"], 70)
Intervals.identify_intervals()
all_intervals = Intervals.return_intervals()

# Identifying the intervals in the activity by heart rate
Intervals = IntervalIdentificationByHeartrate(activity["timestamps"], activity["altitudes"], activity["heartrates"])
Intervals.identify_intervals()
all_intervals = Intervals.return_intervals()

Parsing of Historical weather data from an external service

from sport_activities_features import WeatherIdentification
from sport_activities_features import TCXFile

# Read TCX file
tcx_file = TCXFile()
tcx_data = tcx_file.read_one_file("path_to_file")

# Configure visual crossing api key
visual_crossing_api_key = "weather_api_key" # https://www.visualcrossing.com/weather-api

# Explanation of elements - https://www.visualcrossing.com/resources/documentation/weather-data/weather-data-documentation/
weather = WeatherIdentification(tcx_data['positions'], tcx_data['timestamps'], visual_crossing_api_key)
weatherlist = weather.get_weather(time_delta=30)
tcx_weather = weather.get_average_weather_data(timestamps=tcx_data['timestamps'],weather=weatherlist)
# Add weather to TCX data
tcx_data.update({'weather':tcx_weather})

The weather list is of the following type:

     [
        {
            "temperature": 14.3,
            "maximum_temperature": 14.3,
            "minimum_temperature": 14.3,
            "wind_chill": null,
            "heat_index": null,
            "solar_radiation": null,
            "precipitation": 0.0,
            "sea_level_pressure": 1021.6,
            "snow_depth": null,
            "wind_speed": 6.9,
            "wind_direction": 129.0,
            "wind_gust": null,
            "visibility": 40.0,
            "cloud_cover": 54.3,
            "relative_humidity": 47.6,
            "dew_point": 3.3,
            "weather_type": "",
            "conditions": "Partially cloudy",
            "date": "2016-04-02T17:26:09+00:00",
            "location": [
                46.079871179535985,
                14.738618675619364
            ],
            "index": 0
        }, ...
    ]

Extraction of integral metrics

import sys
sys.path.append('../')

from sport_activities_features.tcx_manipulation import TCXFile

# Read TCX file
tcx_file = TCXFile()

integral_metrics = tcx_file.extract_integral_metrics("path_to_the_file")

print(integral_metrics)

Weather data extraction

from sport_activities_features.weather_identification import WeatherIdentification
from sport_activities_features.tcx_manipulation import TCXFile

#read TCX file
tcx_file = TCXFile()
tcx_data = tcx_file.read_one_file("path_to_the_file")

#configure visual crossing api key
visual_crossing_api_key = "API_KEY" # https://www.visualcrossing.com/weather-api

#return weather objects
weather = WeatherIdentification(tcx_data['positions'], tcx_data['timestamps'], visual_crossing_api_key)
weatherlist = weather.get_weather()

Using with Overpass queried Open Street Map nodes

import overpy
from sport_activities_features.overpy_node_manipulation import OverpyNodesReader

# External service Overpass API (https://wiki.openstreetmap.org/wiki/Overpass_API) (can be self hosted)
overpass_api = "https://lz4.overpass-api.de/api/interpreter"

# External service Open Elevation API (https://api.open-elevation.com/api/v1/lookup) (can be self hosted)
open_elevation_api = "https://api.open-elevation.com/api/v1/lookup"

# OSM Way (https://wiki.openstreetmap.org/wiki/Way)
open_street_map_way = 164477980

overpass_api = overpy.Overpass(url=overpass_api)

# Get an example Overpass way
q = f"""(way({open_street_map_way});<;);out geom;"""
query = overpass_api.query(q)

# Get nodes of an Overpass way
nodes = query.ways[0].get_nodes(resolve_missing=True)

# Extract basic data from nodes (you can later on use Hill Identification and Hill Data Extraction on them)
overpy_reader = OverpyNodesReader(open_elevation_api=open_elevation_api)
# Returns {
#         'positions': positions, 'altitudes': altitudes, 'distances': distances, 'total_distance': total_distance
#         }
data = overpy_reader.read_nodes(nodes)

Extraction of data inside area

import numpy as np
import sys
sys.path.append('../')

from sport_activities_features.area_identification import AreaIdentification
from sport_activities_features.tcx_manipulation import TCXFile

# Reading the TCX file.
tcx_file = TCXFile()
activity = tcx_file.read_one_file('path_to_the_data')

# Converting the read data to arrays.
positions = np.array([*activity['positions']])
distances = np.array([*activity['distances']])
timestamps = np.array([*activity['timestamps']])
heartrates = np.array([*activity['heartrates']])

# Area coordinates should be given in clockwise orientation in the form of 3D array (number_of_hulls, hull_coordinates, 2).
# Holes in area are permitted.
area_coordinates = np.array([[[10, 10], [10, 50], [50, 50], [50, 10]],
                             [[19, 19], [19, 21], [21, 21], [21, 19]]])

# Extracting the data inside the given area.
area = AreaIdentification(positions, distances, timestamps, heartrates, area_coordinates)
area.identify_points_in_area()
area_data = area.extract_data_in_area()

Identify interruptions

from sport_activities_features.interruptions.interruption_processor import InterruptionProcessor
from sport_activities_features.tcx_manipulation import TCXFile

"""
Identify interruption events from a TCX or GPX file.
"""

# read TCX file (also works with GPX files)
tcx_file = TCXFile()
tcx_data = tcx_file.read_one_file("path_to_the_data")

"""
Time interval = time before and after the start of an event
Min speed = Threshold speed to trigger an event / interruption (trigger when under min_speed)
overpass_api_url = Set to something self hosted, or use public instance from https://wiki.openstreetmap.org/wiki/Overpass_API
"""
interruptionProcessor = InterruptionProcessor(time_interval=60, min_speed=2,
                                              overpass_api_url="url_to_overpass_api")

"""
If classify is set to true, also discover if interruptions are close to intersections. Returns a list of [ExerciseEvent]
"""
events = interruptionProcessor.events(tcx_data, True)

Overpy (Overpass API) node manipulation

Generate TCXFile parsed like data object from overpy.Node objects

import overpy
from sport_activities_features.overpy_node_manipulation import OverpyNodesReader


# External service Overpass API (https://wiki.openstreetmap.org/wiki/Overpass_API) (can be self hosted)
overpass_api = "https://lz4.overpass-api.de/api/interpreter"

# External service Open Elevation API (https://api.open-elevation.com/api/v1/lookup) (can be self hosted)
open_elevation_api = "https://api.open-elevation.com/api/v1/lookup"

# OSM Way (https://wiki.openstreetmap.org/wiki/Way)
open_street_map_way = 164477980

overpass_api = overpy.Overpass(url=overpass_api)

# Get an example Overpass way
q = f"""(way({open_street_map_way});<;);out geom;"""
query = overpass_api.query(q)

# Get nodes of an Overpass way
nodes = query.ways[0].get_nodes(resolve_missing=True)

# Extract basic data from nodes (you can later on use Hill Identification and Hill Data Extraction on them)
overpy_reader = OverpyNodesReader(open_elevation_api=open_elevation_api)
# Returns {
#         'positions': positions, 'altitudes': altitudes, 'distances': distances, 'total_distance': total_distance
#         }
data = overpy_reader.read_nodes(nodes)

Missing elevation data extraction

from sport_activities_features import ElevationIdentification
from sport_activities_features import TCXFile

tcx_file = TCXFile()
tcx_data = tcx_file.read_one_file('path_to_file')

elevations = ElevationIdentification(tcx_data['positions'])
"""Adds tcx_data['elevation'] = eg. [124, 21, 412] for each position"""
tcx_data.update({'elevations':elevations})

Example of a visualization of the area detection

Example of a visualization of dead end identification

License

This package is distributed under the MIT License. This license can be found online at http://www.opensource.org/licenses/MIT.

Disclaimer

This framework is provided as-is, and there are no guarantees that it fits your purposes or that it is bug-free. Use it at your own risk!

Cite us

I. Jr. Fister, L. Lukač, A. Rajšp, I. Fister, L. Pečnik and D. Fister, "A minimalistic toolbox for extracting features from sport activity files", 2021 IEEE 25th International Conference on Intelligent Engineering Systems (INES), 2021, pp. 121-126, doi: 10.1109/INES52918.2021.9512927.