quantlab.model.predefined.rnn.ALSTMNet

class quantlab.model.predefined.rnn.ALSTMNet(num_features, num_labels, hidden_size=64, num_layers=2, dropout=0.0, rnn_type='GRU')

Bases: Module

Qlib’s ALSTMModel: a recurrent network with attention over its time steps.

Each bar goes through Tanh(Linear(F, H)) and the recurrent network. A two-layer scorer (Linear(H, H/2), dropout, Tanh, Linear(H/2, 1), softmax over the N bars) weights the hidden states, and Linear(2H, L) maps the last hidden state concatenated with the weighted sum to the output. The parameter names are Qlib’s, so a Qlib state dict loads into a one-label net.

Parameters:

num_featuresint

Features per bar, F.

num_labelsint

Outputs per sample, L; Qlib’s model has one.

hidden_sizeint, default 64

H.

num_layersint, default 2

Recurrent layers.

dropoutfloat, default 0.0

Dropout between recurrent layers and in the scorer.

rnn_type{“GRU”, “LSTM”}, default “GRU”

The recurrent network.

Raises:

ValueError

If rnn_type is neither "LSTM" nor "GRU".

Examples

>>> net = ALSTMNet(num_features=4, num_labels=2, hidden_size=8)
>>> net(torch.randn(5, 3, 4)).shape
torch.Size([5, 2])

forward(x)

Map [B, N, F] windows to [B, L] outputs.

Examples

>>> net.forward(torch.randn(2, 6, 4)).shape
torch.Size([2, 2])