When I use the LMF to predict the slopes hk(x) for new data, the prediction is the same for all samples, which should differ for different x.