##################################################### # Copyright (c) Xuanyi Dong [GitHub D-X-Y], 2019.01 # ##################################################### # I write this package to make AutoDL-Projects to be compatible with the old GDAS projects. # Ideally, this package will be merged into lib/models/cell_infers in future. # Currently, this package is used to reproduce the results in GDAS (Searching for A Robust Neural Architecture in Four GPU Hours, CVPR 2019). ################################################## import torch def obtain_nas_infer_model(config): if config.arch == 'dxys': from .DXYs import CifarNet, ImageNet, Networks genotype = Networks[config.genotype] if config.dataset == 'cifar': return CifarNet(config.ichannel, config.layers, config.stem_multi, config.auxiliary, genotype, config.class_num) elif config.dataset == 'imagenet': return ImageNet(config.ichannel, config.layers, config.auxiliary, genotype, config.class_num) else: raise ValueError('invalid dataset : {:}'.format(config.dataset)) else: raise ValueError('invalid nas arch type : {:}'.format(config.arch))