Crossformer Documentation

To use Crossformer, ensure you have the following Python environment installed:

• Python 3.7.10
• numpy==1.20.3
• pandas==1.3.2
• torch==1.8.1
• einops==0.4.1

To run experiments on standard datasets (like ETTh1), place the datasets in the datasets/ folder and execute main_crossformer.py. The model will automatically train and test, saving checkpoints to checkpoints/ and evaluation metrics to results/.

python main_crossformer.py --data ETTh1 --in_len 168 --out_len 24 --seg_len 6 --itr 1

To use your own data (e.g., AirQualityUCI.csv):

1. Format the data: The CSV should have the first column as a date (or blank) and subsequent columns as the multivariate time series dimensions. Place the file in the datasets/ folder.
2. Train: Run main_crossformer.py specifying the --data name, --data_path, and --data_dim (number of dimensions).
3. Evaluate: Run eval_crossformer.py with the corresponding --setting_name.

# Training with custom data
python main_crossformer.py --data AirQuality --data_path AirQualityUCI.csv --data_dim 13 --in_len 168 --out_len 24 --seg_len 6

# Evaluating custom data
python eval_crossformer.py --setting_name Crossformer_AirQuality_il168_ol24_sl6_win2_fa10_dm256_nh4_el3_itr0 --save_pred

Use eval_crossformer.py to evaluate a specific model checkpoint. You must provide the --checkpoint_root and the exact --setting_name used during training.

python eval_crossformer.py --checkpoint_root ./checkpoints --setting_name Crossformer_ETTh1_il168_ol24_sl6_win2_fa10_dm256_nh4_el3_itr0