iTransformer Lab

Repository for Statistical Learning laboratory on iTransformer. This is a shortened version of the original repository focusing only on the implementation of the iTransformer architecture.

Dataset

In the folder dataset you can find 4 .csv files, which are 4 variants of the of the Electricity Transformer Temperature dataset (ETT).

Description

Datasets contain observations of two Electricity Transformers at two different stations in China. Depending on the variant, you will find two different granularities of observations:

• hourly observations for ETTh1.csv and ETTh2.csv
• minute by minute observations for ETTm1.csv and ETTm2.csv

In each file, the observed variables are:

date	HUFL	HULL	MUFL	MULL	LUFL	LULL	OT
timestamp of observation	High Useful Load	High Useless Load	Middle Useful Load	Middle Useless Load	Low Useful Load	Low Useless Load	Oil Temperature (Target)

For hourly-level files there are a total of 17,420 observations, while for minute by minute files thre are 69,680 observations.

Setup

1. Install miniconda
2. Create and activate a conda environment:

   conda create -n itransformer python==3.11 -y
   conda activate itransformer

3. Install requirements:

   pip install -r requirements.txt

Experiments

In the folder ./scripts you can find a series of bash scripts for performing experiments.

Launch bash scripts

In order to launch an experiment, run the following commands from the terminal in the main direcory:

chmod +x scripts/<script_to_launch>.sh # you may need to give permissions to the file for being executed
bash scripts/<script_to_launch>.sh

For example, suppose you want to run an experiment on the variant h1 of the ETT dataset:

chmod +x scripts/iTransformers_ETTh1.sh
bash ./scripts/iTransformers_ETTh1.sh

Launch with Python

If you want to run one single experiment on a dataset, you can also run directly the python script run.py tuning the configurations to pass as arguments. For example, in order to run the same epxperiment as above:

python -u run.py \
  --is_training 1 \
  --root_path ./dataset/ \
  --data_path ETTh1.csv \
  --model_id ETTh1_96_96 \
  --model iTransformer \
  --data ETTh1 \
  --features M \
  --seq_len 192 \
  --pred_len 96 \
  --e_layers 2 \
  --des 'Exp' \
  --d_model 256 \
  --d_ff 256 \
  --itr 1

The list of all the arguments and their meanings can be found within the script run.py.