Risk Index probably incorrect about simglucose HOT 3 OPEN

Thanks for raising this issue.

The simple add-up comes from the original UVa/Padova Simulator, but I agree that your proposed weighted average makes more sense, and it will bring the final risk index into the same scale of LBGI and HBGI. If you are interested, feel free to submit a PR to the repo, otherwise, I will do that.

FYI, there is a related risk index computation error reported here #62, which is due to the numpy version compatibility, and there is already a PR to fix this computation error.

from simglucose.

This is a follow up to my previous solution. It still suffers from a problem: the risk values are not capped to [0, 100] as they should be. This stems from the assumption in the paper that the bg values should be within [20, 600] mg/dL. Outside of that range we cannot depend on natural log to behave.

Here is the updated implementation - this should directly replace the contents of risk.py:

import numpy as np

def risk_index(BG, horizon):
    # BG is in mg/dL
    BG_to_compute = BG[-horizon:]
    risks =[risk(r) for r in BG_to_compute]
    LBGI = np.mean([r[0] for r in risks])
    HBGI = np.mean([r[1] for r in risks])
    RI = np.mean([r[2] for r in risks])

    return (LBGI, HBGI, RI)

def risk(BG):
    MIN_BG = 20.0
    MAX_BG = 600.0
    if BG <= MIN_BG: 
        return (100.0, 0.0, 100.0)
    if BG >= MAX_BG:
        return (0.0, 100.0, 100.0)
    
    U = 1.509 * (np.log(BG)**1.084 - 5.381)

    ri = 10 * U**2

    rl, rh = 0.0, 0.0
    if U <= 0:
        rl = ri
    if U >= 0:
        rh = ri
    return (rl, rh, ri)

from simglucose.

Opened a PR with the proposed changes. Closed the old PR.

from simglucose.