CTRL

This repository contains the implementation for the paper An NCDE-based Framework for Universal Representation Learning of Time Series.

Requirements

The recommended requirements for CTRL are specified as follows:

• Python 3.8
• torch==1.8.1
• scipy==1.6.1
• numpy==1.19.2
• pandas==1.0.1
• scikit_learn==0.24.2
• statsmodels==0.12.2
• Bottleneck==1.3.2
• torchcde
• h5py

The dependencies can be installed by:

pip install -r requirements.txt

Data

The datasets can be obtained and put into datasets/ folder in the following way:

• UCR datasets should be put into datasets/UCR/ so that each data file can be located by datasets/UCR/<dataset_name>/<dataset_name>_*.csv.

• UEA datasets should be put into datasets/UEA/ so that each data file can be located by datasets/UEA/<dataset_name>/<dataset_name>_*.arff.

• Exchange-rate dataset (obtained from LSTNet repository) placed at datasets/exchange-rate.h5.

• Wind dataset (obtained from https://www.kaggle.com/sohier/30-years-of-european-wind-generation.) placed at datasets/wind.h5.

• ILI dataset (obtained from Autoformer repository ) placed at datasets/ILI.csv.

• Weather dataset (link from Informer repository) placed at datasets/WTH.csv.

• ETT dataset placed at datasets/ETT*.csv.

Usage

To train and evaluate CTRL on a dataset, run the following command:

python train_mul_times.py <dataset_name> <run_name> --loader <loader> --batch-size <batch_size>  --repr-dims <repr_dims>   --eval --taskW <taskW> --hard-neg <hard_neg> --debiase --threshold <threshold> --topk <topk> --max-train-length <len>  --gpu <gpu> --runs <runs>

The detailed descriptions about the arguments are as following:

Parameter name	Description of parameter
dataset_name	The dataset name
run_name	The folder name used to save model, output and evaluation metrics. This can be set to any word
loader	The data loader used to load the experimental data. This can be set to UCR, UEA, forecast_csv, forecast_hdf, imputation.
batch_size	The batch size (defaults to 128)
repr_dims	The representation dimensions (defaults to 320)
eval	Whether to perform evaluation after training
taskW	The trade-off between contrastive task and reconstruction task.
hard-neg	The methods to construct hard negative samples. This can be set to shuffle4, shuffle8, mix2, mix4 or None.
debiase	Whether to eliminate false negative samples
threshold	The similarity threshold to filter false negatives
topk	Proportion of the topk to screen the false negative samples
max-train-length	For sequence with a length greater than <max_train_length>, it would be cropped into some sequences, each of which has a length less than <max_train_length> (defaults to 201)
gpu	The gpu no. used for training and inference (defaults to 0)
runs	Number of executions (defaults to 5)

(For descriptions of more arguments, run python train_mul_times.py -h.)

After training and evaluation, the trained encoder, output and evaluation metrics can be found in training/DatasetName/RunName_Date_Time/.

Scripts: To train and evaluate CTRL on the datasets included in our paper, you can run the script from the ./scripts folder or directly run the python scripts.

For example:

python train_mul_times.py PowerCons  ctrl --loader  UCR --batch-size 128  --max-threads 8   --iters 100 --eval --taskW 0.1 --hard-neg mix4 --debiase --threshold 0.99 --topk 0.2 --runs 5

python train_mul_times.py exchange-rate  ctrl --loader  forecast_hdf --batch-size 128  --max-threads 8   --iters 100 --eval --taskW 0.1 --hard-neg shuffle8 --debiase --threshold 0.98 --topk 0.4 --runs 5

for ratio in $(seq 0.125 0.125 0.5); do
    echo $ratio
    python train_mul_times.py ETTm1  ctrl${ratio} --loader  imputation  --batch-size 128  --max-threads 8   --eval --taskW 0.1 --hard-neg shuffle4 --debiase --threshold 0.98 --topk 0.4 --max-train-length 96 --irregular ${ratio} --runs 5 --iters 200
done