22bd62194e97151e6354edd4d04a3c414083cc9c,model.py,DQN,forward,#DQN#Any#,16

Before Change

    x = self.relu(self.conv3(x))
    x = x.view(x.size(0), -1)
    x = self.relu(self.fc1(x))
    v = self.fc_v(x)
    a = self.fc_a(x)
    return v.expand_as(a) + a - a.mean(1, keepdim=True).expand_as(a)

After Change

    x = self.relu(self.conv3(x))
    x = x.view(x.size(0), -1)
    x = self.relu(self.fc1(x))
    p = torch.stack([self.softmax(z).clamp(min=1e-8, max=1 - 1e-8) for z in self.fc_z(x).chunk(self.action_space, 1)], 1)
    return p  // Probabilities with action over second dimension
    // v = self.fc_v(x)
    // a = self.fc_a(x)
    // return v.expand_as(a) + a - a.mean(1, keepdim=True).expand_as(a)

Instances