functions.py 文件源码

def predictiveQQ(simulations, targets, bands):
    with warnings.catch_warnings():
        warnings.simplefilter("ignore")
    bands = toCustomLogSpace(np.array(bands)[::-1])
    pValues = np.empty_like(targets)
    for i0 in range(pValues.shape[0]):
        sims, idxs = np.unique(simulations[i0,:],return_index=True)
        try:
            pValues[i0] = interp1d(sims, bands[idxs], kind='linear', assume_sorted=True)(targets[i0])
        except np.linalg.linalg.LinAlgError as ex:
            pValues[i0] = np.nan
        except ValueError as ex:
            # TODO: handle better extrapolations
            if targets[i0]<sims[0]:
                pValues[i0] = bands[0]+(bands[0]-bands[1])/(sims[0]-sims[1])*(targets[i0]-sims[0])
            else:
                pValues[i0] = bands[-1]+(bands[-1]-bands[-2])/(sims[-1]-sims[-2])*(targets[i0]-sims[-1])
    pValues = fromCustomLogSpace(pValues)
    pValues[pValues<0] = 0
    pValues[pValues>1] = 1

    pValues = np.sort(1-pValues[np.logical_not(np.isnan(pValues))])
    return (np.linspace(0,1, pValues.shape[0]), pValues)