Generate crappy products and reviews using Amazon's dataset
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
images
prices
productNames
reviews
util
README.md
divinations.json
index.html
sample.lua
sorcery.py

README.md

Crappy Product Generator

This code is the end result of a blog post, training a neural network on Amazon's lowest reviewed products and one star reviews. It uses char-rnn and torch-gan to generate text and images, respectively.

Requirements

You'll need to have working installations of both char-rnn and torch-gan. You'll also need to have Python3 installed, in order to run the script that generates all the outputs you load.

Usage

The socery.py file loads up all the text based char-rnn generators, and uses them to create a new set of products / reviews / prices. It uses the sample.lua script from the char-rnn repository, and includes the utils directory provided with that code. The sorcery.py assumes you've already trained and cleaned a set of images from the torch-gan separately. Refer to the blog post for the directory / image structure it assumes.

Then, to generate a new list of 50 products:

$ python3 sorcery.py 50

You'll notice that there will be times that the neural network won't generate a cohesive, finished review, and so the loop just keeps trying to generate product reviews until it gets finished.

Once that's done, you can view the generated projects in a web browser. Using Python3:

$ python3 -m http.server

Then, open your web browser to http://localhost:8000, and you should be able to see all of your generated products! The sky's the limit!

gan trained on amazon crappy productsgan trained on amazon crappy productsgan trained on amazon crappy products

License

MIT