Spaces:
Sleeping
Sleeping
import torch | |
from torch import nn | |
class ImprovedGRUModel(nn.Module): | |
def __init__(self, | |
input_size=1080, | |
hidden_size=240, | |
output_size=24, | |
num_layers=2, | |
bidirectional=True, | |
dropout_rate=0.1): | |
super(ImprovedGRUModel, self).__init__() | |
self.hidden_size = hidden_size | |
self.num_directions = 2 if bidirectional else 1 | |
self.gru = nn.GRU( | |
input_size=input_size, | |
hidden_size=self.hidden_size, | |
num_layers=num_layers, | |
batch_first=True, | |
dropout=dropout_rate if num_layers > 1 else 0, | |
bidirectional=bidirectional | |
) | |
self.fc1 = nn.Linear(hidden_size * self.num_directions, hidden_size) | |
self.dropout = nn.Dropout(dropout_rate) | |
self.fc2 = nn.Linear(hidden_size, output_size) | |
def forward(self, x): | |
gru_out, _ = self.gru(x) | |
fc1_out = self.fc1(gru_out) | |
fc1_out = torch.relu(fc1_out) | |
fc1_out = self.dropout(fc1_out) | |
output = self.fc2(fc1_out) | |
return output |