DL-Assignment-3

Text-Transliteration.

• Hindi to English transliteration by training a LSTM Encoder and 2-layered Decoder with AttentionMechanism over Google Dakshina dataset and tuned the hyper parameters using parallel plots on Wandb.
• Cell Types used for Encoder and Decoder: RNN, LSTM, GRU.

To run and verify the code please follow the following steps:

• First open the jupiter notebook named Assignment-3.ipynb.
• Then execute the Imports section.
• Then Load the data by executing the Loading Dataset section ( We uploaded dataset on GDrive to access it in Google Colabs).
• There are sections for each question and sub-sections for sub-parts, So execute the section for the question to get the output for that question.
• To run the wandb sweeps first execute the wandb integration section and the question section.
• To run the wandb sweeps first execute the wandb integration section and the question section.
• For Question 5 part we are using a font file 'Lohit-Devanagari.ttf' by uploading it to GDrive.