S4M

Usage

Setup

Suppose you have Anaconda installed to setup environment. First, please install packages using provided environment.yaml.

cd s4m
conda env create -f environment.yaml
conda activate S4M

Dataset

We show the example of running S4M on our benchmark dataset weather with missing observations.

Training

The best checkpoints will be saved in checkpoint directory.

python run.py --is_training 1 --root_path ./data/weather/chunk_missing/ --data_path weather_missing.csv --model_id weather_192_96 --model S4M --per_mem_size 5 --en_conv_hidden_size 256 --data custom4 --topK 20 --thres1 0.95 --thres2 0.6 --M 50 --memory_size 200 --momentum 0.99 --short_len 32 --features M --seq_len 192 --pred_len 96 --e_layers 8 --enc_in 21 --dec_in 21 --c_out 21 --des Exp --d_model 256 --learning_rate 0.005 --train_epochs 15 --num_pattn 30 --num_dG 1 --encoder_type 3 --memnet 13 --mem_type 3 --d_ff 256 --itr 1 --mean_type global_mean --fillna mean