Skilldash V1 — Adaptive Reasoning Trainer

This project implements an adaptive reasoning system that dynamically adjusts difficulty levels based on user performance while continuously collecting data for future machine learning.

It supports three cognitive skills:

• Numerical reasoning
• Logical reasoning
• Pattern recognition

Each skill contains multiple procedural task generators that create unlimited questions at different difficulty levels.

---

How it Works (High Level)

1. A task is generated procedurally from one of several subfamilies.
2. A pretrained ML model predicts:
   • Probability the user will answer correctly (p_correct)
   • Expected response time (expected_time)
3. The user answers the question.
4. Difficulty is updated using a rule-based controller that uses correctness, time, and ML predictions.
5. Every attempt and task feature is logged for analysis and future learning.

This creates a closed-loop adaptive tutoring system with data collection for Task ML.

In V1, ML models only predict performance — they do not choose or modify tasks.
Adaptation happens via difficulty updates, not task parameters.

---

What’s Implemented in V1

• Procedural task generation (no fixed question bank)
• Independent adaptive difficulty for each skill
• Detailed user attempt logging (data_log_final.csv)
• Task feature logging (task_feature_ml.csv)
• Session summaries (session_summary.csv)
• ML models for:
  • predicting correctness
  • predicting response time

---

Data Collection

Every session automatically records:

• User responses, timing, and correctness
• Difficulty transitions per skill
• Task-level structural features

New data can be committed after running sessions so models can be retrained over time. Commit new versions of data_log_final.csv and task_feature_ml.csv after running meaningful sessions.

Future Work (V2): Task ML

Instead of only adjusting scalar difficulty, V2 will aim to learn how to adapt task parameters directly, such as:

• Number of objects in logic problems
• Depth of implication chains
• Noise in comparison tasks
• Step size in arithmetic progressions
• Multipliers in multiplicative patterns

V1 exists primarily to generate high-quality training data for this next phase.

---

How to Run

pip install -r requirements.txt
python session.py