How can I generate the x_enc, x_mark_enc, x_dec, x_mark_dec arguments?

[msasen]

I am trying to export the TimesNET model I have trained. However, the forward() function requires four arguments: x_enc, x_mark_enc, x_dec, and x_mark_dec, and I couldn't understand what these arguments are.

If you can provide support on how to generate these arguments, I will prepare complete documentation on how to export the models found in the Time-Series-Library.

Any pointers on this issue would be greatly appreciated. Thank you.

import torch
from models.TimesNet import Model

class Config:
    def __init__(self):
        self.task_name = "long_term_forecast"
        self.is_training = 1
        self.model_id = "test"  # Updated from ETTh1_96_96
        self.model = "Autoformer"  # Updated from TimesNet
        self.data = "ETTh1"
        self.root_path = "./data/ETT/"  # Updated from ./dataset/ETT-small/
        self.data_path = "ETTh1.csv"
        self.features = "M"
        self.target = "OT"
        self.freq = "h"
        self.checkpoints = "./checkpoints/"


        # Forecasting task
        self.seq_len = 96
        self.label_len = 48
        self.pred_len = 96
        self.seasonal_patterns = "Monthly"
        self.inverse = False

        # Inputation task
        self.mask_rate = 0.25

        # Anomaly detection task
        self.anomaly_ratio = 0.25


        self.expand = 2
        self.d_conv = 4
        self.top_k = 5
        self.num_kernels = 6
        self.enc_in = 7
        self.dec_in = 7
        self.c_out = 7
        self.d_model = 512
        self.n_heads = 8
        self.e_layers = 2
        self.d_layers = 1
        self.d_ff = 2048
        self.moving_avg = 25
        self.factor = 1

        self.seg_len = 96

        # Optimization
        self.num_workers = 10
        self.itr = 1
        self.train_epochs = 10
        self.batch_size = 32
        self.patience = 3
        self.learning_rate = 0.0001
        self.des = "test"
        self.loss = "MSE"
        self.lradj = "type1"
        self.use_amp = False


        self.use_gpu = True
        self.gpu = 0
        self.gpu_type = "cuda"
        self.use_multi_gpu = False
        self.devices = "0,1,2,3"

        # De-stationary projector parameters
        self.p_hidden_dims = [128, 128]
        self.p_hidden_layers = 2

        # Metrics (DTW)
        self.use_dtw = False

        # Augmentation
        self.augmentation_ratio = 0
        self.seed = 2
        self.jitter = False
        self.scaling = False
        self.permutation = False
        self.randompermutation = False

        self.magwarp = False
        self.timewarp = False
        self.windowslice = False
        self.windowwarp = False
        self.rotation = False
        self.spawner = False
        self.dtwwarp = False
        self.shapedtwwarp = False
        self.wdba = False
        self.discdtw = False
        self.discsdtw = False
        self.extra_tag = ""
        self.dropout = 0.1
        # TimeXer
        self.patch_len = 16


        self.task_name = "long_term_forecast"
        self.is_training = 1
        self.root_path = "./dataset/ETT-small/"
        self.data_path = "ETTh1.csv"
        self.model_id = "ETTh1_96_96"
        self.model = "TimesNet"
        self.data = "ETTh1"
        self.features = "M"
        self.seq_len = 96
        self.label_len = 48
        self.pred_len = 96
        self.e_layers = 2
        self.d_layers = 1
        self.factor = 3
        self.enc_in = 7
        self.dec_in = 7
        self.c_out = 7
        self.d_model = 16
        self.d_ff = 32
        self.des = "Exp"
        self.itr = 1
        self.top_k = 5
        self.num_kernels = 6
        self.embed = "timeF"
        self.freq
        #self.activation = "relu"
        #self.output_attention = True
        #self.do_predict = True
configs = Config()

checkpoint = torch.load('checkpoints/long_term_forecast_ETTh1_96_96_TimesNet_ETTh1_ftM_sl96_ll48_pl96_dm16_nh8_el2_dl1_df32_expand2_dc4_fc3_ebtimeF_dtTrue_Exp_0/checkpoint.pth')

model = Model(configs)
model.load_state_dict(checkpoint) 
with torch.no_grad():
    predictions = model(...)

print("Predictions:", predictions)

TypeError: forward() missing 4 required positional arguments: 'x_enc', 'x_mark_enc', 'x_dec', and 'x_mark_dec'

[wuhaixu2016]

Hi, thanks for your detailed experiments.

The configuration for these four arguments can be found here:

https://github.com/thuml/Time-Series-Library/blob/main/data_provider/data_loader.py#L92-L103
https://github.com/thuml/Time-Series-Library/blob/main/exp/exp_long_term_forecasting.py#L111-L112

[wuhaixu2016]

Hi, hope your issue has been resolved.

Also, it is very appreciated that you make a pull request for the code for exporting well-trained models.