6. All these factors taken collectively create an LTSM mannequin (LTSM Bundle) that outperforms all current strategies of reprogramming LLM for time collection and transformer-based time collection forecasting fashions.
Reprogram your LTSM your self!
Need to reprogram your individual LTSM? Here is a tutorial for the LTSM bundle: https://github.com/daochenzha/ltsm/blob/main/tutorial/README.md
Step 1: Create a digital setting. Clone and set up the necessities and repositories.
conda create -n ltsm python=3.8.0
conda activate ltsm
git clone git@github.com:daochenzha/ltsm.git
cd ltsm
pip3 set up -e .
pip3 set up -r necessities.txt
Step 2: Put together the dataset: Be sure your native knowledge folder appears to be like like this:
- ltsm/
- datasets/
DATA_1.csv/
DATA_2.csv/
DATA_3.csv/
...
Step 3: Generate time collection prompts from the coaching, validation, and check datasets
python3 prompt_generate_split.py
Step 4: Discover the generated time collection prompts within the ‘./prompt_data_split’ folder. Then run the next command to finalize the prompts.
# normalizing the prompts
python3 prompt_normalization_split.py --mode match#export the prompts to the "./prompt_data_normalize_split" folder
python3 prompt_normalization_split.py --mode rework
Remaining step: Practice your individual LTSM utilizing Time Sequence Immediate and Linear Tokenization with gpt2-medium.
python3 main_ltsm.py
--model LTSM
--model_name_or_path gpt2-medium
--train_epochs 500
--batch_size 10
--pred_len 96
--data_path "DATA_1.csv DATA_2.csv"
--test_data_path_list "DATA_3.csv"
--prompt_data_path "prompt_bank/prompt_data_normalize_split"
--freeze 0
--learning_rate 1e-3
--downsample_rate 20
--output_dir [Your_Output_Path]
For extra info, see our paper and GitHub repository.
paper: https://arxiv.org/pdf/2406.14045
code: https://github.com/daochenzha/ltsm/
reference:
[1] Liu, Pengfei, et al. “Pretraining, Prompting, and Predicting: A Systematic Investigation of Prompting Strategies in Pure Language Processing.” ACM Computing Survey 55.9(2023):1–35.
[2] Liu, Xiao, et al., “Self-supervised studying: generative or contrastive?” IEEE Transactions on Information and Information Engineering. 35.1(2021):857–876.
[3] Ansari, Abdul Fatir, et al., “Chronos: Studying the Language of Time Sequence.” arXiv preprint arXiv:2403.07815 (2024).
