TsFormer

TsFormer is a toolbox that implement transformer models on Time series data

Todo

1. data preprocess

• gefcom2014
• uci
• ETT

2. dataloader

• dataloaders

3. models

• rnn
• lstm
• GRU
• ESN
• CNN
• TCN
• transformer
• informer
• autoformer

4. TODO

• Spacetimeformer
• SCINet
• Deep learning for time series forecasting
• PyTorch Forecasting
• tsai
• flow-forecast

5. train and evaluate

6. Custom Informer

python -u run_autoformer.py \
        --is_training 1 \
        --root_path ./data/electricity/ \
        --data_path electricity.csv \
        --model_id ECL \
        --model informer \
        --data custom \
        --features S \
        --seq_len 96 \
        --label_len 48 \
        --pred_len 96 \
        --e_layers 2 \
        --d_layers 1 \
        --factor 3 \
        --enc_in 1 \
        --dec_in 1 \
        --c_out 1 \
        --embed fixed \
        --des 'Exp' \
        --itr 1


mse:0.2755982279777527, mae:0.3857262134552002,rmse:0.524974524974823, mape:1.9572646617889404, mspe:238.20448303222656

python -u run_autoformer.py \
        --is_training 1 \
        --root_path ./data/electricity/ \
        --data_path electricity.csv \
        --model_id ECL \
        --model informer \
        --data custom \
        --features S \
        --seq_len 96 \
        --label_len 48 \
        --pred_len 96 \
        --e_layers 2 \
        --d_layers 1 \
        --factor 3 \
        --enc_in 1 \
        --dec_in 1 \
        --c_out 1 \
        --embed timeF \
        --des 'Exp' \
        --itr 1

mse:0.22287048399448395, mae:0.3356129825115204,rmse:0.4720916152000427, mape:1.6913783550262451, mspe:260.3700866699219

Transformer Results

python -u run_autoformer.py \
        --is_training 1 \
        --root_path ./data/electricity/ \
        --data_path electricity.csv \
        --model_id ECL \
        --model transformer \
        --data custom \
        --features S \
        --seq_len 96 \
        --label_len 48 \
        --pred_len 96 \
        --e_layers 2 \
        --d_layers 1 \
        --factor 3 \
        --enc_in 1 \
        --dec_in 1 \
        --c_out 1 \
        --embed timeF \
        --des 'Exp' \
        --itr


mse:0.284598171710968, mae:0.38772597908973694,rmse:0.5334774255752563, mape:2.1156060695648193, mspe:381.4866943359375

AutoFormer

python -u run_autoformer.py \
        --is_training 1 \
        --root_path ./data/electricity/ \
        --data_path electricity.csv \
        --model_id ECL \
        --model autoformer \
        --data custom \
        --features M \
        --seq_len 96 \
        --label_len 48 \
        --pred_len 96 \
        --e_layers 2 \
        --d_layers 1 \
        --factor 3 \
        --enc_in 321 \
        --dec_in 321 \
        --c_out 321 \
        --des 'Exp' \
        --itr 1

mse:0.2043592780828476, mae:0.3170555830001831,rmse:0.45206114649772644, mape:3.2521157264709473, mspe:414847.125