Skip to content

Latest commit

 

History

History
85 lines (58 loc) · 2.33 KB

README.md

File metadata and controls

85 lines (58 loc) · 2.33 KB
Raw

AIs deserve love too

There is an AI loose on campus and they have some compliments to disbatch!

Step 1 - Scrape image URLS

  1. rename creds.example.ts to creds.ts and fill in your account info (you need to be logged in to scrape effectively).
  2. Run the following
cd Step1
yarn install
npx parcel build index.ts -t node
node dist
  1. You will be left with a Step1.json file containing the post URLs in the root directory

Step 2 - Download images

  1. Make sure you have run step 1
  2. Run these commands
cd Step2
python download_images.py

Step 3 - Extract text

First we need to install the GCloud SDK with

brew install --cask google-cloud-sdk

then login to CLI and setup project

gcloud auth login
gcloud config set project PROJECTID

now run gcloud ml vision detect-text Step2/images/1.jpg and make sure its working. This may ask you to enable Vision API.

Convert your images to connections.txt with

cd Step3
sh ExtractTexts.sh

Step 4 - Scrape UVM student data

  1. rename creds.example.ts to creds.ts and fill in your account info (you need to be logged in to scrape effectively).
  2. Run the following
cd Step4
yarn install
npx parcel build index.ts -t node
node dist
  1. You will be left with a acc_handles.json file containing scraped instagram handles in the Step4 directory.

Step 5 - OpenAI's GPT 3 - Generate connections

OpenAI does not allow for their models to be used on social media unfortunately, so we will need to recreate this using a different technique.

Step 5 - Fine-tuning GPT-2 - Generate connections

Upload FineTuneLoveBot.ipynb to Google Colab and follow the steps in there. When it has finished running, copy models and checkpoint folders from Google Drive into Step5-EasyGPT folder. Install OpenCV, gpt_2_simple, and starlette with a python package manager of your choice.

Run python server.py and wait until it starts listening. You will now me able to generate connections by sending GET requests to http://localhost:8080/?prefix=NAME_GOES_HERE. Note: these will take ~30 seconds to generate.

Step 6 - Post connections

With Step 5's server still running and Step 4's acc-handles.json generated, and creds.example.ts configured like we had done above, run

yarn
yarn build
node dist

and watch LoveBot get to work.