Sequence2Sequence model to translate date strings into the standard date format.
Dec 10, 1865 --> 1865-12-10
December 25, 1991 --> 1991-12-25
Saturday June 10 1741 --> 1741-06-10
This is for educational purposes to learn about seq2seq models and attention mechanisms.
- Helpful primer with cool visuals
- Bahdanau et al.
- Luong et al.
- Conceptual implementation guide. ML Mastery
- Vaswani et al.
A good Seq2Seq tutorial can be found here.
- Install conda.
- Create and activate conda env
conda create -n date python=3.7
conda activate date
- Install python packages.
pip install -r requirements.txt
- Activate the conda env.
conda activate date
- Start your jupyter notebook.
jupyter notebook
- Generate the dataset by running notebook
00_generate_data.ipynb
. - Complete the Seq2Seq w/ attention code in
01_FITB_seq2seq.ipynb
. A solution is found in01_seq2seq.ipynb
. - Classify written dates w/ a transformer architecture.
02_transformers.ipynb
.
- Feel free to experiment with changing how the data is generated. You can widen or narrow the date range, or come up with different ways to represent dates as strings.