Along with the continued development and the addition of new sensors and features to the RAPIDS pipeline, tests for the currently available sensors and features are being implemented. Since this is a Work In Progress this page will be updated with the list of sensors and features for which testing is available. For each of the sensors listed a description of the data used for testing (test cases) are outline. Currently for all intent and testing purposes the `tests/data/raw/test01/` contains all the test data files for testing android data formats and `tests/data/raw/test02/` contains all the test data files for testing iOS data formats. It follows that the expected (verified output) are contained in the `tests/data/processed/test01/` and `tests/data/processed/test02/` for Android and iOS respectively. `tests/data/raw/test03/` and `tests/data/raw/test04/` contain data files for testing empty raw data files for android and iOS respectively.
The following is a list of the sensors that testing is currently available.
Due to the difference in the format of the raw battery data for iOS and Android as well as versions of iOS the following is the expected results the `battery_deltas.csv`. This would give a better idea of the use cases being tested since the `battery_deltas.csv` would make both the iOS and Android data comparable. These files are used to calculate the features for the battery sensor.
- The battery delta data file contains data for 1 day.
- The battery delta data contains 1 record each for a `charging` and
`discharging` episode that falls within an `epoch` for every
`epoch`. Thus, for the `daily` epoch there would be multiple
`charging` and `discharging` episodes
- Since either a `charging` episode or a `discharging` episode and
not both can occur across epochs, in order to test episodes that
occur across epochs alternating episodes of `charging` and
`discharging` episodes that fall across `night` to `morning`,
`morning` to `afternoon` and finally `afternoon` to `night` are
present in the battery delta data. This starts with a
`discharging` episode that begins in `night` and end in `morning`.
- There is one battery data file each, for testing both iOS and
Android data formats.
- There is also an additional empty data file for both android and
There are two wifi features (`phone wifi connected` and `phone wifi visible`). The raw test data are seperatly stored in the `phone_wifi_connected_raw.csv` and `phone_wifi_visible_raw.csv`.
- One episode for each daily segment (`night`, `morning`, `afternoon` and `evening`) on each day
- Two episodes near the transition of the daily segment, one starts at the end of the afternoon, `Fri 17:10:00` and another one starts at the beginning of the evening, `Fri 18:01:00`
- One episode across two segments, `daily` and `30-mins`, (from `Fri 05:55:00` to `Fri 06:00:41`)
- Two episodes locate in the same daily segment (`Sat 12:45:36` and `Sat 16:48:22`)
- One episode before the time switch, `Sun 00:15:06`, and one episode after the time switch, `Sun 06:01:00`
- The raw keyboard data file contains data for 4 days.
- The raw keyboard data contains records with difference in `timestamp` ranging from
milliseconds to seconds.
- With difference in timestamps between consecutive records more than 5 seconds helps us to create separate
sessions within the usage of the same app. This helps to verify the case where sessions have to be different.
- The raw keyboard data contains records where the difference in text is less
than 5 seconds which makes it into 1 session but because of difference of app
new session starts. This edge case determines the behaviour within particular app
and also within 5 seconds.
- The raw keyboard data also contains the records where length of `current_text` varies between consecutive rows. This helps us to tests on the cases where input text is entered by auto-suggested
or auto-correct operations.
- One three-minute episode with a 1-minute row on Sun 08:59:54.65 and 09:00:00,another on Sun 12:01:02 that are considering a single episode in multi-timezone event segments to showcase how
inferring time zone data for Keyboard from phone data can produce inaccurate results around the tz change. This happens because the device was on LA time until 11:59 and switched to NY time at 12pm, in terms of actual time 09 am LA and 12 pm NY represent the same moment in time so 09:00 LA and 12:01 NY are consecutive minutes.
- The feature requires raw application foreground data file and raw phone screen data file
- The raw data files contains data for 4 day.
- The raw conversation data contains records with difference in `timestamp` ranging from milliseconds to minutes.
- An app episode starts when an app is launched and ends when another app is launched, marking the episode end of the first one,
or when the screen locks. Thus, we are taking into account the screen unlock episodes.
- There are multiple apps usage within each screen unlock episode to verify creation of different app episodes in each
screen unlock session. In the screen unlock episode starting from Fri 05:56:51, Fri 10:00:24, Sat 17:48:01, Sun 22:02:00, and Mon 21:05:00 we have multiple apps, both system and non-system apps, to check this.
- The 22 minute chunk starting from Fri 10:03:56 checks app episodes for system apps only.
- The screen unlock episode starting from Mon 21:05:00 and Sat 17:48:01 checks if the screen lock marks the end of episode for that particular app which was launched a few milliseconds to 8 mins before the screen lock.
- Finally, since application foreground is only for Android devices, this feature is also for Android devices only. All other files are empty data files
- A one-minute sedentary episode on Sun 02:00:00. It exists in November but not in February in STZ
- A five-minute sedentary episode on Fri 11:58:00. It is split within two 30-min segments and the morning
- A three-minute lightly active episode on Fri 11:10:00, a one-minute at 11:18:00 and a one-minute 11:24:00. These check for start and end times of first/last/longest episode
- A three-minute fairly active episode on Fri 11:40:00, a one-minute at 11:48:00 and a one-minute 11:54:00. These check for start and end times of first/last/longest episode
- A three-minute very active episode on Fri 12:10:00, a one-minute at 12:18:00 and a one-minute 12:24:00. These check for start and end times of first/last/longest episode
- A eight-minute MVPA episode with intertwined fairly and very active rows on Fri 12:30:00
- The above episodes contain six higmet (>= 3 MET) episodes and nine lowmet episodes.
- One two-minute sedentary episode with a 1-minute row on Sun 09:00:00 and another on Sun 12:01:01 that are considering a single episode in multi-timezone event segments to showcase how inferring time zone data for Fitbit from phone data can produce inaccurate results around the tz change. This happens because the device was on LA time until 11:59 and switched to NY time at 12pm, in terms of actual time 09 am LA and 12 pm NY represent the same moment in time so 09:00 LA and 12:01 NY are consecutive minutes.
- A three-minute sedentary episode on Sat 08:59 that will be ignored for multi-timezone event segments.
- A three-minute sedentary episode on Sat 12:59 of which the first minute will be ignored for multi-timezone event segments since the test segment starts at 13:00
- A three-minute sedentary episode on Sat 16:00
- A four-minute sedentary episode on Sun 10:01 that will be ignored for Novembers's multi-timezone event segments since the test segment ends at 10am on that weekend.
- A three-minute very active episode on Sat 16:03. This episode and the one at 16:00 are counted as one for lowmet episodes
- A main sleep episode that starts on Fri 20:00:00 and ends on Sat 02:00:00. This episode starts after 11am (Last Night End) which will be considered as today's (Fri) data.
- A nap that starts on Sat 04:00:00 and ends on Sat 06:00:00. This episode starts before 11am (Last Night End) which will be considered as yesterday's (Fri) data.
- A nap that starts on Sat 13:00:00 and ends on Sat 15:00:00. This episode starts after 11am (Last Night End) which will be considered as today's (Sat) data.
- A main sleep that starts on Sun 01:00:00 and ends on Sun 12:00:00. This episode starts before 11am (Last Night End) which will be considered as yesterday's (Sat) data.
- A main sleep that starts on Sun 23:00:00 and ends on Mon 07:00:00. This episode starts after 11am (Last Night End) which will be considered as today's (Sun) data.
- Any segment shorter than one day will be ignored for sleep RAPIDS features.
Checklist
|time segment| single tz | multi tz|platform|
|-|-|-|-|
|30min|OK|OK|fitbit|
|morning|OK|OK|fitbit|
|daily|OK|OK|fitbit|
|threeday|OK|OK|fitbit|
|weekend|OK|OK|fitbit|
|beforeMarchEvent|OK|OK|fitbit|
|beforeNovemberEvent|OK|OK|fitbit|
## Fitbit Sleep Intraday
Description
- A five-minute main sleep episode with asleep-classic level on Fri 11:00:00.
- An eight-hour main sleep episode on Fri 17:00:00. It is split into 2 parts for daily segment: a seven-hour sleep episode on Fri 17:00:00 and an one-hour sleep episode on Sat 00:00:00.
- A two-hour nap on Sat 01:00:00 that will be ignored for main sleep features.
- An one-hour nap on Sat 13:00:00 that will be ignored for main sleep features.
- An eight-hour main sleep episode on Sat 22:00:00. This episode ends on Sun 08:00:00 (NY) for March and Sun 06:00:00 (NY) for Novembers due to daylight savings. It will be considered for `beforeMarchEvent` segment and ignored for `beforeNovemberEvent` segment.
- A nine-hour main sleep episode on Sun 11:00:00. Start time will be assigned as NY time zone and converted to 14:00:00.
- A seven-hour main sleep episode on Mon 06:00:00. This episode will be split into two parts: a five-hour sleep episode on Mon 06:00:00 and a two-hour sleep episode on Mon 11:00:00. The first part will be discarded as it is before 11am (Last Night End)
- Any segment shorter than one day will be ignored for sleep PRICE features.
