Hello,

I am trying to reproduce the results for ETTh1 dataset.

When running the following file in Colab: !python -u train.py --data ETTh1 --input_len 240 --output_len 24 --seq_len 48 --dec_seq_len 48 --pred_len 24 --features S --iterations 10 --exps 5 --hidden_size 96 --n_heads 3 --n_encoder_layers 3 --encoder_attention full --n_decoder_layers 3 --decoder_attention full --batch_size 32 --embedding_size 24 --dropout 0.1

I run into the following error: File "train.py", line 189, in main() File "train.py", line 185, in main preform_experiment(args) File "train.py", line 155, in preform_experiment preds, trues = run_iteration(deepspeed_engine if args.deepspeed else model , train_loader, args, training=True, message=' Run {:>3}, iteration: {:>3}: '.format(args.run_num, iter)) File "train.py", line 109, in run_iteration result = model(batch) File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 727, in _call_impl result = self.forward(*input, **kwargs) File "/content/drive/MyDrive/Query/model.py", line 287, in forward e = self.encs0 File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 727, in _call_impl result = self.forward(*input, **kwargs) File "/content/drive/MyDrive/Query/model.py", line 239, in forward return super(Linear, self).forward(x) File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/linear.py", line 93, in forward return F.linear(input, self.weight, self.bias) File "/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py", line 1692, in linear output = input.matmul(weight.t()) RuntimeError: mat1 dim 1 must match mat2 dim 0

Can you please help me resolve this issue?, thank you.

best wishes Janamejaya

Hi again, and thanks for the implementation of the 'ms' feature.

Two questions:

1. Is it possible to only inversely transform the numbers of a prediction? Running inverse transform or disabling scaling on an entire training explodes the gradient with no effort. In short; is it possible to only do an inverse transformation on the pred_len?

2. With 'ms' enabled, how would one expand on it to predict three outputs instead of a singular, is it as simple as feeding a list to the target, which is now defined as 'OT' ?

I hope you can help out, best regards, and again thanks for your contribution!

Hi i'm trying to implement the code, but with 'MS' feature, which i can see traces of in your code, inspired from the Informer.

I tried adding a 'multiuni' prediction type, alongside the 'multi' and 'uni' one and have changed my setting accordingly to use 'multiuni':

elif self.prediction_type == 'multiuni':
            sys.argv.extend(["--features", 'MS'])
            sys.argv.extend(["--input_len", '7', "--output_len", "1"])

I am however, consistently getting issues, most recently, triggering an assertion in the code:

File "train.py", line 178, in preform_experiment
    len(train_data.data_y[0]), args.output_len)
AssertionError: Dataset contains output vectors of length 7 while output_len is set to 1

Is there anyway that you could guide me to the implementation of MS?

Best regards

Hi, just a quick question.

I didnt't have a chance to run the code, so I assume the input data have a shape like B,L,C or B,C,L. Line279: self.enc_input_fc = Linear(input_size, dim_val). I think “input_size” means the length while “dim_val” is somewhat related to embedding dimensions. And the shape of the tensor turned into a square which doesn't make sense to me. Is that a typo or I misunderstand something? Many thanks!

Hello, Thank you for releasing the code, I had couple of questions. I am trying to run the code in Google Colab.

This is my training script !python train.py --data='ETTm1' --seq_len=720 --pred_len=24 --dec_seq_len=24 --hidden_size=128 --batch_size=100 --embedding_size=32 --n_encoder_layers=3 --encoder_attention='full' --n_decoder_layers=3 --decoder_attention='full' --dropout=0.1 --iterations=10 --exps=5 --features='M' --n_heads=3 --input_len=7 --output_len=7

After 1 iteration I am getting the following error :

Traceback (most recent call last): File "train.py", line 190, in main() File "train.py", line 186, in main preform_experiment(args) File "train.py", line 159, in preform_experiment ipc.sendPartials(iter, mse, mae) File "/content/drive/MyDrive/query-selector-master/ipc.py", line 59, in sendPartials c.connect(('localhost', PORT)) ConnectionRefusedError: [Errno 111] Connection refused

Please do let me know, Thanks.

Best Regards Niharika

Hello,

I had a doubt in the data processing step

1. In ETTm2 dataset in data_loader to split the dataset we use the following code to calculate border and then calculate the number of train and test points. Can you explain this line of code? How is this calculated ? border1s = [0, 12*30*24*4 - self.seq_len, 12*30*24*4+4*30*24*4 - self.seq_len] border2s = [12*30*24*4, 12*30*24*4+4*30*24*4, 12*30*24*4+8*30*24*4]

2. seq_len is calculated as 24 * 4 * 4 how is this calculated? What does it signify?

Please let me know

Thank you Niharika

Hello Jacek,

I was going through the query selector paper, and I am trying to reproduce your code. I have a few doubts :

1. Where is the train, test, and validation split on the dataset done?

2. In train.py we train the model, and then validation is done. Where is testing happening?

3. If our pred length is 24, we get 24 predictions after the last data point, So how can we save the values predicted by the model? In which part of the code is happening ?

Please do let me know

Thank you Regards Niharika Joshi

Hello, your work is excellent. However, when I used the electricity dataset (https://github.com/laiguokun/multivariate-time-series-data), a memory explosion problem occurred. I don't know if there is a solution. When running the EETh data set, it is normal. Thank You.