In this project, I worked to addressing the problem of generating a new Seinfeld TV scripts using RNNs. Given a word, the RNN model will automatically predict the most likely next word. In order to generate a text of words, firstly we need to determine the start word and specific length of words to repeat the prediction. Then, we will obtain a text which represent a new “fake” TV script. I did this project as a part of Deep Learning Nanodegree from Udacity. I did this project as a part of Deep Learning Nanodegree from Udacity.
The data is a subset of Seinfeld TV dataset of scripts from 9 seasons. You can find more information for the full dataset on Kaggle.
The high-level steps of the project include:
- Preprocessing of dataset (include lookup table, tokenization)
- Prepare the batches
- Build the RNN network. It consists of embedding layer, LSTM layer and fully connected layer.
- Training and testing.
- Create TV script generation algorithm – this algorithm takes the model I created, start word, length of text to generate new text.
I test the model with jerry as start word and 400 total length of text , the results are showed below.
jerry:!(waves the keys from jerry)
jerry: oh.
elaine:(to kramer) you know, i got a problem with the woman.
jerry:(to elaine) what?
george: i don't want to know that.
jerry: what is wrong with you, that is so bad.
george: what are you doing here, jerry, you know? you know, i have no idea how to have sex with the guy....
jerry: oh! yeah, you know, i think i'm gonna get out of here.
george:(to george) you know, you know, i think i could get you a little more enthusiastic...
george:(pointing to the counter) oh, hi, jerry.
george: what about my last date?!
kramer:(shouts) i don't think you can take a bite!
jerry: i thought you said that.
jerry: i thought you were going to be a lot of a good time, i don't have any trouble. i was a little bit of the commercial.
kramer: yeah, i know.
jerry:(to jerry) i don't know. i mean, if you could just do it.
george:(sarcastic) i thought he was a kid. i can't do this!
george:(to george) hey, what are you doing?
george: i think i can. i just don't think we have a good thing.
kramer: well, i don't know what to do.
george: well, if we should just get a little tired of that.
jerry: oh, i don't think so. i think it's a good idea. i mean, i have to go to the bathroom.
jerry: i don't know why you can't do it.
kramer: i don't think so.
george: i don't think so.
elaine: oh.... no. no, no no no
- Python 3.7
- Numpy
- Torch
- Jupyter Notebook
- Maybe needs to use GPU
The whole project is located in the jupyter notebook file dlnd_tv_script_generation, you can use the Anaconda environment to open the Jupyter Notebook and install the requirement.