zhiyongc/graph_convolutional_lstm

Traffic Graph Convolutional Recurrent Neural Network

graph_convolutional_lstm's Issues

Missing import?

Hi,

I'm trying out your code to gain a better understanding of the whole concept, but I get an error:

---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-21-5955e675779a> in <module>
     2 back_length = 3
     3 Clamp_A = False
----> 4 gclstm = GraphConvolutionalLSTM(K, torch.Tensor(A), FFR[back_length], A.shape[0], Clamp_A=Clamp_A, output_last = True)
     5 gclstm, gclstm_loss = TrainModel(gclstm, train_dataloader, valid_dataloader, num_epochs = 1)
     6 gclstm_test = TestModel(gclstm, test_dataloader, max_speed )

<ipython-input-17-b87e6355c7f7> in __init__(self, K, A, FFR, feature_size, Clamp_A, output_last)
    28
    29         # a length adjustable Module List for hosting all graph convolutions
---> 30         self.gc_list = nn.ModuleList([FilterLinear(feature_size, feature_size, self.A_list[i], bias=False) for i in range(K)])
    31
    32         hidden_size = self.feature_size

<ipython-input-17-b87e6355c7f7> in <listcomp>(.0)
    28
    29         # a length adjustable Module List for hosting all graph convolutions
---> 30         self.gc_list = nn.ModuleList([FilterLinear(feature_size, feature_size, self.A_list[i], bias=False) for i in range(K)])
    31
    32         hidden_size = self.feature_size

NameError: name 'FilterLinear' is not defined

FilterLinear is not something you defined, and it's not something you import either. Do you remember where it comes from?

Obviously, since it's been a while since you posted your code I am not running the same version of PyTorch (1.12.1+cu116 instead of 0.42), but it doesn't seem like it's an import as you're not doing any import star.

Predicted vs Actual Speed Discrepancy

@zhiyongc

With your code I was able to obtain the validation loss graph which looks similar to yours (10^-4) for the Seattle Loop Detector dataset. But when I tried to do a comparison between actual vs predicted, it's not even close to what is shown in your paper. Can you tell me if you are doing something differently or you are using a different code altogether?

Below is the predicted vs actual speed graph for your reference.

Thanks,
Suhas

About FFR matrix

if i have a raw data，how can i get the FFR matrix？

The Seattle data link could not be opened

Hello, I'm sorry to bother you, but I can't open the Seattle data link you gave me, so I hope you can share the Seattle detector data again. Thank you

IndexError

File "/data/home/u19120777/Graph_Convolutional_LSTM-master/Code_V1/Train_Validate.py", line 108, in TrainRNN
loss_interval_train = np.around(sum(losses_train[-interval:]).cpu().numpy()[0]/interval, decimals=8)
IndexError: too many indices for array: array is 0-dimensional, but 1 were indexed

my pytorch version is 1.1 python version is 1.7 and I run this model with gpu, but I get troubled with this error..

How to run your code

Hello,
I tried running your code but it was not successful.
python Main.py
Here is the detail error:

Epoch 0/0
----------
Traceback (most recent call last):
  File "Main.py", line 93, in <module>
    rnn, rnn_loss = TrainRNN(train_dataloader, valid_dataloader, num_epochs = 1)
  File "/home/tiendh/Projects/IoT++/Graph_Convolutional_LSTM/Train_Validate.py", line 72, in TrainRNN
    loss_train = loss_MSE(outputs, labels)
  File "/home/tiendh/deeplearning_env/lib/python3.5/site-packages/torch/nn/modules/module.py", line 477, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/tiendh/deeplearning_env/lib/python3.5/site-packages/torch/nn/modules/loss.py", line 421, in forward
    return F.mse_loss(input, target, reduction=self.reduction)
  File "/home/tiendh/deeplearning_env/lib/python3.5/site-packages/torch/nn/functional.py", line 1716, in mse_loss
    return _pointwise_loss(lambda a, b: (a - b) ** 2, torch._C._nn.mse_loss, input, target, reduction)
  File "/home/tiendh/deeplearning_env/lib/python3.5/site-packages/torch/nn/functional.py", line 1674, in _pointwise_loss
    return lambd_optimized(input, target, reduction)
RuntimeError: input and target shapes do not match: input [40 x 323], target [40 x 1 x 323] at /pytorch/aten/src/THCUNN/generic/MSECriterion.cu:12

Actually, I had to enable the line train_dataloader, valid_dataloader, test_dataloader, max_speed = PrepareDataset(speed_matrix) to pass the very first error.
Am I doing something wrong? Please guide me how to run your code.