09

Comet for Natural Language Processing

The following example performs sentiment analysis of Disneyland Reviews Dataset, available on Kaggle at this link and released under the CC0: Public Domain license.

The dataset contains 42,656 reviews of 3 Disneyland branches (- Paris, California and Hong Kong), posted by visitors on Trip Advisor. For each review, it also provides the rating, which ranges from 1 (totally unsatisfied) to 5 (satisfied). We group ratings into two categories: positive, if the rating is greater than 2, and negative, otherwise. The example compares a pretrained model and a custom model, and shows the results in Comet.

Requirements

• You need to create a .comet.config file and place it in this directory:

[comet]
api_key=YOUR_COMET_API_KEY
workspace=YOUR_WORKSPACE
project_name=YOUR_PROJECT_NAME

• You need to install the packages contained in the requirements.txt file. To install them, you can runn the following command:

pip install -r requirements.txt

Results in Comet

You can see the results in Comet at the following links:

• the dashboard
• the report