README.md

metadata

language:
  - en
license: mit
tags:
  - xgboost
  - sentence-transformers
  - text-classification
  - tabular
metrics:
  - accuracy
  - f1
pipeline_tag: text-classification

Difficulty Classifier Model (XGBoost + Sentence Transformers)

This repository contains an XGBoost classification model designed to categorize question difficulty into three levels: Easy, Medium, and Hard.

Architecture & Pipeline

1. Text Enrichment: Inputs are structured by combining the core question_content, any multiple choice options (formatted as a pipe-separated string), and associated passage_text context.
2. Vector Space Embeddings: The enriched question string is converted into a 384-dimensional dense vector using the sentence-transformers/all-MiniLM-L6-v2 embedding engine.
3. Tabular Classifier: An XGBoost Classifier (XGBClassifier) predicts the difficulty class from the embedding vector.

Model Files

• difficulty_xgb_local_model.joblib: The trained XGBoost model binary.
• difficulty_label_encoder.joblib: Scikit-Learn LabelEncoder mapping classes to integers (['Easy', 'Hard', 'Medium']).

How to Use

You can load and use these assets directly using joblib and sentence-transformers:

import joblib
import json
import pandas as pd
import numpy as np
from sentence_transformers import SentenceTransformer

# 1. Load the model, label encoder and embedder
model = joblib.load("difficulty_xgb_local_model.joblib")
label_encoder = joblib.load("difficulty_label_encoder.joblib")
embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")

# 2. Define single item payload
question = "The number of species having non-pyramidal shape is..."
options = ["CO3 2-", "SO3", "NO3-"]

# 3. Format inputs (match training pipeline schema)
options_text = " | ".join(options)
enriched_text = f"Question: {question} \n Options: {options_text}"

# 4. Generate embeddings and run inference
embedding = embedder.encode([enriched_text])
prediction_idx = model.predict(embedding)[0]
predicted_label = label_encoder.classes_[prediction_idx]

print("Predicted Difficulty:", predicted_label)