Create ActiGraph counts from raw data

Since ActiGraph has released their count-processing algorithm, it is now fairly easy to obtain ActiGraph/ActiLife counts even without using ActiLife. In this tutorial, we will show how to do this using PAAT.

%%capture
import paat

Load the data

First, we have to load our example GT3X file again:

data, sample_freq = paat.read_gt3x("data/10min_recording.gt3x")
data

	X	Y	Z
2022-01-03 10:20:00.000	0.804688	0.621094	0.085938
2022-01-03 10:20:00.010	0.804688	0.597656	0.085938
2022-01-03 10:20:00.020	0.804688	0.585938	0.078125
2022-01-03 10:20:00.030	0.804688	0.582031	0.074219
2022-01-03 10:20:00.040	0.800781	0.585938	0.074219
...	...	...	...
2022-01-03 10:29:59.950	0.289062	0.960938	-0.050781
2022-01-03 10:29:59.960	0.289062	0.960938	-0.054688
2022-01-03 10:29:59.970	0.285156	0.957031	-0.054688
2022-01-03 10:29:59.980	0.289062	0.957031	-0.054688
2022-01-03 10:29:59.990	0.285156	0.960938	-0.050781

60000 rows × 3 columns

Create ActiGraph counts over arbitrary epoch lengths

PAAT has implemented a calculate_actigraph_counts() functions which requires the data, the sample_freq and a string indicating the epoch length (e.g. "1s" for one second epochs, "10s for ten second epochs or "1min" for one minute epochs). In the backend, PAAT is passing most of the processing to ActiGraph’s agcounts which could also be used independently. However, we implemented a convenience wrapper that allows passing the epoch length as a string (as commonly done with pandas) and which correctly updates the DataFrame’s index with the correct time stamps.

Here, for example is the calculation of one second epochs shown:

counts_1sec = paat.calculate_actigraph_counts(data, sample_freq, "1s")
counts_1sec.head()

	Y	X	Z
2022-01-03 10:20:00	0	0	0
2022-01-03 10:20:01	0	0	0
2022-01-03 10:20:02	0	0	0
2022-01-03 10:20:03	0	0	0
2022-01-03 10:20:04	0	0	0

To calculate ten second epochs, one does only need to adjust the epoch_length argument.

counts_10sec = paat.calculate_actigraph_counts(data, sample_freq, "10s")
counts_10sec.head()

	Y	X	Z
2022-01-03 10:20:00	0	0	0
2022-01-03 10:20:10	0	0	0
2022-01-03 10:20:20	0	0	0
2022-01-03 10:20:30	0	0	0
2022-01-03 10:20:40	185	376	204

If you created a one second epoch DataFrame, it is also no problem to aggregate it to arbitrary epochs later using Pandas’ resample function:

new_counts_10sec = counts_1sec.resample("10s").sum()
new_counts_10sec.head()

	Y	X	Z
2022-01-03 10:20:00	0	0	0
2022-01-03 10:20:10	0	0	0
2022-01-03 10:20:20	0	0	0
2022-01-03 10:20:30	0	0	0
2022-01-03 10:20:40	185	376	204

Create Brønd counts

Alternatively, you can use the open source algorithm proposed by Brønd et al. (2017):

brond_counts = paat.calculate_brond_counts(data, sample_freq, "10s")
brond_counts.head()

	Y	X	Z
2022-01-03 10:20:00	28	35	0
2022-01-03 10:20:10	0	0	0
2022-01-03 10:20:20	0	0	0
2022-01-03 10:20:30	0	0	0
2022-01-03 10:20:40	188	380	205

An alternative implementation can be found here.