Skip to content

Code to run the ExtRA algorithm for unsupervised topic/aspect extraction on English texts.

License

Notifications You must be signed in to change notification settings

wayfair-incubator/extra-model

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Code style: black Imports: isort GitHub license PyPI

Table of Contents

extra-model

Code to run the Extra algorithm for the unsupervised topic/aspect extraction on English texts.

Read the Official Documentation here

Quick start

IMPORTANT:

  1. When running Extra inside docker-container, make sure that Docker process has enough resources. For example, on Mac/Windows it should have at least 8 Gb of RAM available to it. Read More about RAM Requirements
  2. GitHub repo does not come with Glove Embeddings. See section Downloading Embeddings for how to download the required embeddings.

Using docker-compose

This is a preferred way to run extra-model. You can find instructions on how to run extra-model using CLI or as a Python package here

First, build the image:

docker-compose build

Then, run following command to make sure that extra-model was installed correctly:

docker-compose run test

Downloading Embeddings

Next step is to download the embeddings (we use Glove from Stanford in this project).

To download the required embeddings, run the following command:

docker-compose run --rm setup

The embeddings will be downloaded, unzipped and formatted into a space-efficient format. Files will be saved in the embeddings/ directory in the root of the project directory. If the process fails, it can be safely restarted. If you want to restart the process with new files, delete all files except README.md in the embeddings/ directory.

[Optional] Run docker-compose build again

After you've downloaded the embeddings, you may want to run docker-compose build again. This will build an image with embeddings already present inside the image.

The tradeoff here is that the image will be much bigger, but you won't spend ~2 minutes each time you run extra-model waiting for embeddings to be mounted into the container. On the other hand, building an image with embeddings in the context will increase build time from ~3 minutes to ~10 minutes.

Run extra-model

Finally, running extra-model is as simple as:

docker-compose run extra-model /package/tests/resources/100_comments.csv

NOTE: when using this approach, input file should be mounted inside the container. By default, everything from extra-model folder will be mounted to /package/ folder. This can be changed in docker-compose.yaml

This will produce a result.csv file in /io/ (default setting) folder.

There are multiple flags you can set to change input/outputs of extra. You can find them by running:

docker-compose run extra-model --help

Learn more

Our official documentation is the best place to continue learning about extra-model:

  1. Explanation of inputs/outputs
  2. Step-by-step workflow of what happens inside of extra-model
  3. Examples of how extra-model can be used in downstream applications
  4. Detailed explanation of how to run extra-model using different interfaces (via docker-compose, via CLI, as a Python package).

Authors

extra-model was written by mbalyasin@wayfair.com, mmozer@wayfair.com.