Methods and apparatus for predicting wireless measurements based on virtual access points is described. In some embodiments, a location of a user equipment (UE) may be obtained, and given the location of the UE, an output may be generated using a machine learning model, the output including one or more predicted wireless measurements. The output may be indicative of a wireless channel in a multipath environment. In some variants, the machine learning model may have been trained by obtaining a training dataset including multipath components data and ground truth locations of a wireless device, and performing an optimization with respect to the multipath components data and the ground truth locations. In some implementations, a training output including a predicted multipath component may be produced during the training, and the optimization may include an iterative minimization of an error between at least the predicted multipath component and a labeled multipath component.