hihihihiwsf / ast Goto Github PK

Hi @hihihihiwsf ! Congratulations for your very interesting paper and thank you for sharing the code ! I am trying to run your code, could you please provide an example of your params.json file ?

Hi, Thanks for your code sharing. Could you please attach a corresponding requirement file for the necessary libraries?

Could you please indicate that you used code from https://github.com/zhykoties/TimeSeries for preprocess_data.py?

def test(model, params, x, v_batch, id_batch):
batch_size = x.shape[0]
sample_mu = torch.zeros(batch_size, params.predict_steps, device=params.device)
sample_q90 = torch.zeros(batch_size, params.predict_steps, device=params.device)
src_mask, memory = model.encode(x[:, :params.predict_start,:], id_batch)
for t in range(params.predict_steps):
ys = x[:, params.predict_start:params.predict_start+t+1,:]
out = model.decode(memory, ys, id_batch, src_mask)
q50, q90 = model.generator(out)
if t!=0:
q50 = q50[:, -1]
q90 = q90[:, -1]
sample_mu[:, t] = q50 * v_batch[:, 0] + v_batch[:, 1]
sample_q90[:, t] = q90* v_batch[:, 0]
if t < (params.predict_steps - 1):
x[:, params.predict_steps+t+1, 0] = q50
return sample_mu, sample_q90

1. In the loop "for t in range(params.predict_steps):", the loop executes once and returns?
2. When forecasting, the value of the prediction range is unknown. Why is ys = x[:, params.predict_start:params.predict_start+t+1,:] is used as the input of the decoder.
3. if t < (params.predict_steps - 1): x[:, params.predict_steps+t+1, 0] = q50 What does this step mean?

I wonder if it is possible to share your code in github?

In this lines,The self.q90 tensor could not do backwards，How can it to update the parameters?

class Generator(nn.Module):
    def __init__(self, params):
        super(Generator, self).__init__()
        self.q50 = nn.Linear(params.d_model, 1)
        self.q90 = nn.Linear(params.d_model, 1)

Hi, many thanks for releasing this code. I am wondering whether you would be able to upload any code to reproduce the experiments from the paper? In particular, it would be great to see how you built vanilla transformer and sparse transformer, then evaluated their performance against AST. Thanks again!

hello, I'm trying to run the code, and there is some error about
"RuntimeError: The size of tensor a (160) must match the size of tensor b (24) at non-singleton dimension 2"
at gan_transformer.py at line 181, in forward
return self.a_2 * (x - mean) / (std + self.eps) + self.b_2
so I want to kown if the code is wrong.
looking forward to your reply, thanks.

Hi, thanks for your code sharing. I tried to run the training process, but I am stuck in many Ipython sections, there is no description about how did you guys use Ipython in the code running process. Could you share this parts also?

Can you send me the elect data set? I can't download the elect data set.Thank you very much!
Thanks a million!
Email: [email protected]