You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, avg_delta in glucose-get-last.js by taking the current bg, subtracting a past bg value, up to 3 steps behind (15 min old), and dividing by the minutes between. A problem with this approach is that avg_delta is about as noisy as delta. Any decisions based on avg_delta are bound to be corrupted by noise. A better approach is to use an FIR filter over past n samples (this could be a variable) to estimate an average slope. Here is an example that compares delta FIR filtered avg_delta using the current bg and past 6 values, so averaging is over 30 minutes. This FIR filter is equivalent to finding the slope of the best-fit linear-regression line through the bg points. Should investigate other FIR filters.
The text was updated successfully, but these errors were encountered:
Currently, avg_delta in glucose-get-last.js by taking the current bg, subtracting a past bg value, up to 3 steps behind (15 min old), and dividing by the minutes between. A problem with this approach is that avg_delta is about as noisy as delta. Any decisions based on avg_delta are bound to be corrupted by noise. A better approach is to use an FIR filter over past n samples (this could be a variable) to estimate an average slope. Here is an example that compares delta FIR filtered avg_delta using the current bg and past 6 values, so averaging is over 30 minutes. This FIR filter is equivalent to finding the slope of the best-fit linear-regression line through the bg points. Should investigate other FIR filters.
The text was updated successfully, but these errors were encountered: