Summary

This package contains the official implementation of Towards Machine Learning on data from Professional Cyclists. It is built to predict the heart rate of professional cyclists given time series of measurements collected by bike computers like srm or pioneer devices. The current configuration gives the algorithm feedback of the heart rate from 30 time steps ago to allow it to model the cyclists physiological response to work.

Installation

The recommended method of installation is to first clone the repository and install it in edit mode, to allow modifying the configuration to your usage:

git clone [email protected]:agrinh/procyclist_performance.git
pip install -e procyclist_performance

Usage

Run a training session with:

python -m procyclist_performance.train

Modify which parameters to extract in dataset.py. The first training will produce a cache of the preprocessed data which is re-used on subsequent runs, until the load function is modified. The input and target paramters are chosen in main of train.py.

Data

Concepts

The package reads collections of data or Sessions. This makes it possible to use it separately for e.g. different groups of cyclists in the team. Each collection of Sessions may contain multiple riders, each with a number of training or racing sessions. Furthermore, since data is often collected by different devices, each configured device should have a glob pattern to match data produced by that particular device.

Configuration

The package is configured by setting the configuration files in (procyclist/config)[procyclist/config]. Configuring:

• Collection of Sessions : sessions.cfg
• Devices used to collect data: devices.cfg

Collection of Sessions

Follow the examples in sessions.cfg to add new collections. The DEFAULT section (which applies to all collections) defines:

• Name of matrix in each <cyclist name>.mat file containing metadata about cyclist (meta_matrix)
• Relative path from package to file specifying names for each column in the metadata matrix (meta_parameters)
• Path to use for storing preprocessed data (cache_path)

By default the package has the following data collections (Sessions) configured:

• example: /data/example
• men: /data/men
• women: /data/women

Devices

The devices.cfg comes with settings for SRM and Pioneer devices (though these must be altered to match how you handle your data). Specified for each device is:

• Name of matrix in each <session>.mat file with session information (matrix)
• Relative path from package to file specifying names for each column in the matrix (parameters)
• A filename pattern which should match all filenames produced by the device ('filename')

Specify collection / device

Once the collections and devices are specified, configure which ones to use for training in procyclist/dataset.py. E.g.:

sessions = Sessions.create(
  name='collection-name',
  device='device-name',
  ...
)

Structure

In the root directory of a collection of Sessions there should be directories corresponding to cyclist names. Each should contain a metadata .mat file with the riders name, and a data directory with .mat files for each cycling session. These cycling session files may carry any name, but should have a name matching the device pattern string.

Example

Below is an example directory structure compatible with this package and the Session collection example (defined in sessions.cfg) and device pioneer (defined in devices.cfg).

/data/example
├── cyclist_1
│   ├── cyclist_1.mat
│   └── data
│         ├── session_1_pioneer.mat
│         ├── session_2_pioneer.mat
│         └── < ... >
├── cyclist_2
│   ├── cyclist_2.mat
│   └── data
│         ├── session_1_pioneer.mat
│         ├── session_2_pioneer.mat
│         └── < ... >
.
.
.

This configuration expects the cyclist metadata .mat files (cyclist_n.mat) to contain a matrix named alles, with columns specified by config/parameters_alles.csv. Individual sessions (session_n_pioneer.mat) should contain a cell array named data_pioneer, with element {0}{0} being a matrix where rows are timesteps and columns are as specified in config/parameters_alles.csv.

Citations

If you found this package useful in your research please use the following citation for the paper (Towards Machine Learning on data from Professional Cyclists)[https://arxiv.org/abs/1808.00198]:

@inproceedings{hilmkilcycling,
  title={Towards machine learning on data from professional cyclists},
  author={Hilmkil, Agrin and Ivarsson, Oscar and Johansson, Moa and Kuylenstierna, Dan and van Erp, Teun},
  booktitle={Proceedings of the 12th World Congress on Performance Analysis of Sports},
  publisher={Faculty of Kinesiology, University of Zagreb},
  pages={168--176}
  address={Opatija, Croatia},
  year={2018}
}