Neural Date Translation

Date translation from conventional Human readable format to machine readable format(YYYY-MM-DD) using Neural Machine Translation. For example a date in human readable format can be : ‘sunday 15 september 2013’, ‘29-oct-1997’ or ‘30 august 1985’ etc.
The task is to convert this to a more normalized format like YYYY-MM-DD.

For this task a sequence to sequence encoder-decoder network has been used. LSTM cell units have been used in both the encoder as well as the decoder network. This model also use Attention Mechanism.

Alt Text

There are three main files:

  1. Neural Date Translation (Jupyter Notebook) : Contains the main code along with explanation. Also this is to be used is retraining is required.
  2. Neural Date Translation (Python script) : Incase you just want to see this in action. This doesn’t train the model instead uses the saved weight values.
  3. utility : used for generating and preprocessing training data.

Here is a snapshot output for some new test dates.

alt text

Credits:

This project is based on the assignment from Sequence Models Specialization by Deeplearning.ai on Coursera.
https://www.coursera.org/learn/nlp-sequence-models/home/welcome