wadpac / sundararajan-sleepclassification-2021

Classification of sleep stages using accelerometer data

This repository includes the code corresponding to the various experiments and plots from the paper Sundararajan et al. 2021 Sleep classification from wrist-worn accelerometer data using random forests published in Nature Scientific Reports: https://www.nature.com/articles/s41598-020-79217-x.

How to apply the open access model to your data?

For instructions on how to use the trained sleep/wake/non-wear classification model see: https://github.com/wadpac/SleepStageClassification/blob/master/ggir_ext

sundararajan-sleepclassification-2021's Issues

how to run it

Is there a "quickstart" document anywhere, otherwise I'll study the code.

I have XYZ binary raw accel sleep data and want to classify sleep stages (light, deep.., REM).

Results not being written to csv output

Thanks for sharing this great package!

When I run the package in 'binary' mode, I can see the 'wake_sleep' results when I load the RData file out of the 'basic' folder:

However, I do not see the 'wake_sleep' results in the csv file in the 'csv' folder:

Also, when I try to run the package in 'multiclass' mode I get a TypeError: unsupported operand type(s) for /: 'NoneType' and 'float' on line 54 from get_sleep_stage.py ( y_pred = y_pred/float(nfolds) )

Here is the way I am calling 'multiclass', do I need to change the input for 'colnames'?

BUG finding sleep states

There seems to be a bug in how sleeps states are found:
https://github.com/wadpac/SleepStageClassification/blob/5b288995e62bbd66faa66bd932b06af8a65f8445/preprocessing/preproc_psgnewcastle.py#L71

Below is a plot, on the top raw the sleep states derived from get_sleep_states and on the bottom the psg file sleep states. See the red circles, in the get_sleep_states are labeled as NaN, whereas in the original psg they are labeled:

In the middle of a psg recording period there should not be any NaNs as the psg data is always labeled.