How I Got to Top 24% on a Kaggle Text Classification Challenge Without Writing a Single Line of Code

In this post, we will see how to use the platform and get a submission that achieves a respectable 83% Accuracy on the test set.

This accuracy may seem low but the current agenda for the platform is to also finally have API to use these models with as low latency as possible. So currently, the models used strike the perfect balance between accuracy and latency.

Based on the awesome Huggingface Transformers Library.

So let’s start . . .

Aasaan.ai Website

First, we will visit the Aasaan.ai website.

Click on try now to start the demo. It is completely free!

Import the dataset

First, we will download the dataset from the Kaggle Challenge website. 

We are tackling the Disaster Tweets Challenge — https://www.kaggle.com/c/nlp-getting-started/data

Upload the train.csv and wait for a few seconds.

Select the text and the labels

Aasaan.ai shows you a preview of the dataset showing you the columns.

The text column is column 3 and the label column is column 4, so we will select these for training the model.

Train the model

Now we need to just train the model. Yes. It is that easy!

Now we wait. It takes roughly 2 minutes to train the model. Have a coffee.

Create the submission File

Move to the CSV tab.

Upload the test.csv previously downloaded, and wait for a few seconds and the submission.csv will be downloaded.

Upload to Kaggle and view the results. I got to top 24% of all participants!

Interested?

We have successfully built, trained and deployed a data classifier without -

1. Writing a single line of code
2. Configuring any GPU
3. Looking at daunting documentations
4. Reading and understanding of machine learning papers.

Visit us at aasaanai to try for yourself and be an early beta tester and join our waiting list!

Aasaan.ai is a No Code platform for building, training and deploying Deep Learning models without worrying about GPUs, Infrastructure, Transformers or anything.