Activity count feature addition

[simon-p-2000]

Added the option to download, process and read the activity counts from the mesa dataset. This is done by extending the read_mesa function. The activity counts represent processed accelerometer data and can be used as an additional feature to train the existing classifiers.

The option to download and process the activity counts is integrated into the read_mesa function by a variable activity_source and is set to False by default. As such, unless specified, no activity counts will be used. If either cached activity counts or the actigraphy csv files in the mesa database are used, the SleepRecord which the function returns, contains an additional array with the activity counts for the specified recording.

This new feature is integrated in such a way that it works for both the offline and non offline option provided by the read_mesa function. Since the actigraphy csv files are much smaller than the EDF files, no option to delete the csv files afer downloading them is provided.

[cbrnr]

I have made two cosmetic commits (please check if you're OK with the changes).

[simon-p-2000]

Running the ws_gru_mesa classifier without the activity feature leads to this result

[cbrnr]

Yep, this looks good! Can you run the other two as well and compare with https://sleepecg.readthedocs.io/en/stable/classification/?

[simon-p-2000]

wrn_gru_mesa_weighted:

[simon-p-2000]

wrn_gru_mesa classifier

It appears that the classifiers are working similarly to before the addition of the activity count feature, so when the activity counts are not added as a feature the classifier is not affected. The next step would be to run the classifiers again with the activity counts added as a feature and see if that improves the classifier performance.

[cbrnr]

Thanks @simon-p-2000!