Using OpenAI's GPT-2 to revive someone from the past
In loving memory of Kobe, here are some examples:
👀 @DeMar_DeRozan #LegendsLive http://www.youtube.com/watch?v=e5E0Godoe_04
I just knew I loved watching them. Thank you @RealSkipBayless https://twitter.com/realbigzapboy/status/7761841583bolikno8 🔥 🦅
Camp was phenomenal! Amazing people. Thank you for all the support. From the most kids to the highest, kids can't wait to see what #ThePunies can teach them. pic.twitter.com/Br8MgniLO
Congrats @demar_derozan #WIZENOUT
Bravo @DwyaneWade 🙌 #assfwd
🙌 #competition https://twitter.com/devinoremiller/status/845137575970065009 …
The game of @blakegriffin32 has grown tremendously #respect
Badges
- Clone this repo to your local machine using
git clone https://github.com/Skyline-9/GoodbPy.gitOnce inside the project make sure to install all the dependencies (it's just TWINT and it's dependencies)
pip install -r requirements.txtOnce you have TWINT installed, just run setup.sh using
sh setup.shUsing CLI, you can scrape all the tweets of a specific user (basically the person you want to simulate) and save to a text file using
twint -u username -o file.txtOr you can export as a .csv file
twint - u username -o file.csv --csvOnce you have the tweets saved, you have to delete the timestamps for the text file. I used the JetBrains Column Selection mode, but I might write a script to automatically remove that later (which shouldn't be hard, since the extraneous information is always the same number of character for each tweet).
Open up the .ipynb file using Google Colab (or you can run the code locally as well, I just don't have the GPU). Upload the text file to your Google Drive - this is for linking it to the Google Colab so we can use the training data. Run through each cell of the notebook and have fun!
- Scripting the scraping and training part so all you have to do is input a username
- Scrape from more sources (e.g. Facebook posts) so the there is more training data
- Using GANs to create fake faces of the person