Inspiration

from popular examples of RNN generating characters trained on specific dataset like shakespear plays and html generated-code by RNN. inspired from Udacity deep learning with Pytorch

What it does

generate Arabic poems never written before

How I built it

PyTorch RNN netwrok with LTSM cells

Challenges I ran into

1- get a dataset to work on ( i collected it with myself ) 2- sampling inputs to the network and getting right output format 3- working with arabic letters

Accomplishments that I'm proud of

scarbbing Arabic poetry-dataset that contains contain 55K poem for over 540 poet from 19 different countries and 9 different eras at link scrapped entirly from link

What I learned

1-webscarpping 2-RNN 3-LSTM 4-PyTorch workflow

What's next for Arabic poem Generator

1- getting better results using the right sample rather than choosing random sample from the data-set i used. 2- consider what is caller "sea of the poetry" (Bahar) which is the way the poem sounds when we say it, Generally make a more sophisticated poem generator

Share this project:

Updates