Model Fine Tuning

[rajeman]

Week 6 Exercise — Emmanuel

Fine-tuning a model to predict a student's salary package (LPA) based on their academic and skill profile.

Overview

Uses the Student Placement Prediction dataset from Kaggle. The model is fine-tuned via OpenAI's API (GPT-4.1-nano) to predict salary in LPA (lakhs per annum) from student features.

Features & Target

• Features: branch, college_tier, cgpa, coding_skills, aptitude_score, communication_skills, ml_knowledge
• Target: salary_package_lpa

Pipeline

1. Load & clean — Download dataset via kagglehub, normalize columns, handle missing values
2. Split — Train (60%), validation (20%), test (20%)
3. Prepare JSONL — Format as user/assistant message pairs for OpenAI fine-tuning
4. Upload & fine-tune — Upload JSONL to OpenAI, create fine-tuning job (gpt-4.1-nano)
5. Evaluate — Custom evaluator on test set: MAE, RMSE, R², error trend chart, actual vs predicted scatter. Plots saved to plots/