Attention Please!! – Beyond the Loss

Hi there! I’m Ashish, a Data Scientist at Khan Academy. I dabble in applied machine learning, NLP, analytics, and the fine art of storytelling with data. Along the way, I share lessons from my journey.

New here? Start from something below or visit the About page for context.

Solution Approach - Analytics Vidhya November Jobathoon

competition

solution

Documentation of the solution approach for Employee Attrition Prediction Problem in Analytics Vidhya November Jobathon

Authorship Identification: Part-1 (The baseline)

project

NLP

LLM

Given a large number of authors of texts, the goal is to correctly identify the author of a given text. This problem is considered hard especially when less data is available and texts are small (typically with 500 words or less). In this part I have setup BiLSTM model as baseline for the task.

Enable Fira Code and Ligatures in code cells

tutorial

productivity

This tutorial shows how to enable the coding font Fira Code and Ligatures for code cells in your blog post generated with fastpages.

Authorship Identification - Part-2 (DistilBERT Transformer)

project

NLP

LLM

Using transfer-learning to fine-tune pretrained DistilBERT transformer for authorship identification. In a nutshell, DistilBERT is a small version of BERT which is smaller, faster, cheaper, and lighter. It has 40% less parameters original BERT, runs 60% faster and preserve over 95% of BERT’s performances.

Credit Card Lead Prediction - EDA

EDA

Visualization

This project delves into the EDA of Credit Card Lead Prediction dataset. The problem is to predict if an existing customer of a mid-sized Bank is likely to purchase their Credit Card.