Filename convention

Store numerical or categorical feature data derived from sensor or event files.

Filename convention

FeatureType.FEATUREID-VERSIONCODE.YYYY-MM-DD-HH-mm-ss-SSS-[M/P]HHmm.feature.csv

Feature files could be in arbitrary length. But for on-the-fly computation, it is recommended to store features in hourly length to keep the file size reasonably small for fast I/O.

This often ships with source code that indicates the computation of the feature, the source code should be clear about FEATUREID and VERSIONCODE.

FeatureType

1. Used to represent the name of the feature. If multiple features are combined into one file, becoming a feature set, then feature type should clearly state the name of the feature set.

2. CamelStyle, allowed characters: alphabets and digits.

3. Required.

Examples

For single feature file: Correlation, DominantFrequency, PhoneVariance, DominantWristActivityCount.

For feature set file: TimeDomainFeatures, DominantWristFeatures.

FEATUREID

1. Locally unique identifible for current feature set, flexible to include location, id, and any information.

2. UPPERCASE style, allowed characters: alphabets, digits and dash.

3. Required.

Examples

For feature set: DOMINANTWRIST, DOMINANTWRIST-PARTICIPANT1TO3 For single feature: DOMIANTWRIST-CORRELATION, [IMEI]-CORRELATION

VERSIONCODE

1. It's possible feature gets updated, then use VERSIONCODE to indicate the difference.

2. Optional, if omit, use NA instead.

3. Allowed characters: alphabets and digits.

Timestamp

1. Timestamp has the same convention as that of sensor data. The time represents the first row of annotation.