Where can i find documentation about Flux modular (approach) for building models?

Specific Domains Machine Learning
1

i found this example ,but no documentation how to build this way

struct FFNetwork
    fc_1
    dropout
    fc_2
    FFNetwork(
        input_dim::Int, hidden_dim::Int, dropout::Float32, num_classes::Int
    ) = new(
        Dense(input_dim, hidden_dim, relu),
        Dropout(dropout),
        Dense(hidden_dim, num_classes),
    )
end

function (net::FFNetwork)(x)
    x = Flux.flatten(x)
    return net.fc_2(net.dropout(net.fc_1(x)))
end
2

this pytorch code i want to reproduce to Julia Flux

class FFNetwork(Module):
    def __init__(self, input_dims, hidden_dim, dropout_ratio, num_classes):
        super(FFNetwork, self).__init__()
        self.flat_image_dims = np.prod(input_dims)
        self.fc_1 = torch.nn.Linear(self.flat_image_dims, hidden_dim)
        self.dropout = torch.nn.Dropout(dropout_ratio)
        self.fc_2 = torch.nn.Linear(hidden_dim, num_classes)

    def forward(self, x):
        x = x.view(-1, self.flat_image_dims)
        return self.fc_2(self.dropout(F.relu(self.fc_1(x))))
3

Per Advanced Model Building · Flux, all you need to add to make the custom layer Flux compatible is @functor FFNetwork.

3 Likes
4

thanks, but i already figure it out…

5

Great that you already figured it out! In future, I would recommend posting what you found here and marking it as a solution as soon as you’ve done so. It saves us time answering a solved question, gives a solution for future readers and is generally good etiquette.

10 Likes