search

GFNOrg / EB_GFN

Code for our paper "Generative Flow Networks for Discrete Probabilistic Modeling"
MIT License
73 stars 19 forks source link

Which network architectures you used? #4

Open matekrk opened 1 year ago

matekrk commented 1 year ago

Hi, many thanks for the paper and providing the code. In emb I can see:

if args.model.startswith("mlp-"):
        nint = int(args.model.split('-')[1])
        net = network.mlp_ebm(np.prod(args.input_size), nint)
    elif args.model.startswith("cnn-"):
        nint = int(args.model.split('-')[1])
        net = network.MNISTConvNet(nint)
    elif args.model.startswith("resnet-"):
        nint = int(args.model.split('-')[1])
        net = network.ResNetEBM(nint)
    else:

        raise ValueError("invalid model definition")

I wonder if you could share your MNISTConvNet and/or ResNetEBM with the configurations how to train them on MNIST? Thanks!