scripts/process_foundational_assist.py

#!/usr/bin/env python3
"""
FoundationalASSIST Processing Script for DIF Analysis
======================================================
Run AFTER you get access to: https://huggingface.co/datasets/ASSISTments/FoundationalASSIST

This script:
1. Downloads all 3 data files (Interactions, Problems, Skills)
2. Cleans the HTML/MathML from Problem Body to readable text
3. Computes item-level statistics  
4. Merges everything into one DIF-ready table
5. Exports to parquet + CSV

Usage:
    pip install huggingface_hub pandas pyarrow beautifulsoup4 lxml
    python process_foundational_assist.py
"""
import os, json, re, sys
import pandas as pd
from huggingface_hub import hf_hub_download

OUT = "foundational_assist_data"
os.makedirs(OUT, exist_ok=True)
REPO = "ASSISTments/FoundationalASSIST"

# ============================================================
# Step 1: Download
# ============================================================
print("Downloading FoundationalASSIST data files...")
files = {
    "Data/Interactions.csv": "interactions.csv",
    "Data/Problems.csv": "problems.csv", 
    "Data/Skills.csv": "skills.csv",
}

for hf_path, local_name in files.items():
    local_path = os.path.join(OUT, local_name)
    if os.path.exists(local_path):
        print(f"  Already exists: {local_name}")
    else:
        try:
            path = hf_hub_download(repo_id=REPO, filename=hf_path, repo_type="dataset")
            import shutil
            shutil.copy(path, local_path)
            print(f"  ✅ Downloaded: {local_name} ({os.path.getsize(local_path)/1e6:.1f} MB)")
        except Exception as e:
            print(f"  ❌ {local_name}: {e}")
            if "GatedRepoError" in str(type(e).__name__) or "403" in str(e):
                print("\n⚠️  ACCESS DENIED — You need to accept the data agreement at:")
                print("    https://huggingface.co/datasets/ASSISTments/FoundationalASSIST")
                print("    Then re-run this script.")
                sys.exit(1)

# ============================================================
# Step 2: Load data
# ============================================================
print("\nLoading data...")
interactions = pd.read_csv(os.path.join(OUT, "interactions.csv"))
problems = pd.read_csv(os.path.join(OUT, "problems.csv"))
skills = pd.read_csv(os.path.join(OUT, "skills.csv"))

print(f"Interactions: {len(interactions):,} rows")
print(f"  Columns: {list(interactions.columns)}")
print(f"Problems: {len(problems):,} rows")  
print(f"  Columns: {list(problems.columns)}")
print(f"Skills: {len(skills):,} rows")
print(f"  Columns: {list(skills.columns)}")

# ============================================================
# Step 3: Clean HTML from Problem Body
# ============================================================
print("\nCleaning Problem Body HTML → plain text...")

try:
    from bs4 import BeautifulSoup
except ImportError:
    os.system("pip install beautifulsoup4 lxml -q")
    from bs4 import BeautifulSoup

def clean_html(html_str):
    """Convert HTML/MathML problem body to readable text.
    Based on the approach in the dataset's Code/cleantext.py"""
    if pd.isna(html_str) or not isinstance(html_str, str):
        return ""
    
    # Parse HTML
    soup = BeautifulSoup(html_str, "html.parser")
    
    # Handle MathML: extract text content 
    for math_tag in soup.find_all("math"):
        # Try to extract readable math from MathML
        math_text = math_tag.get_text(separator=" ", strip=True)
        math_tag.replace_with(f" [{math_text}] ")
    
    # Handle images (note: images may contain diagrams)
    for img in soup.find_all("img"):
        alt = img.get("alt", "")
        if alt:
            img.replace_with(f"[Image: {alt}]")
        else:
            img.replace_with("[Image]")
    
    # Handle tables
    for table in soup.find_all("table"):
        rows = []
        for tr in table.find_all("tr"):
            cells = [td.get_text(strip=True) for td in tr.find_all(["td", "th"])]
            rows.append(" | ".join(cells))
        table.replace_with("\n".join(rows))
    
    # Get text
    text = soup.get_text(separator=" ", strip=True)
    
    # Clean up whitespace
    text = re.sub(r'\s+', ' ', text).strip()
    
    # Clean up common HTML entities
    text = text.replace('&amp;', '&').replace('&lt;', '<').replace('&gt;', '>')
    text = text.replace('&nbsp;', ' ').replace('&#39;', "'")
    
    return text

# Apply cleaning
problems["problem_text_clean"] = problems["Problem Body"].apply(clean_html)

# Also clean MC options if present
for col in ["Multiple Choice Options", "Multiple Choice Answers", "Fill-in Options", "Fill-in Answers"]:
    if col in problems.columns:
        problems[f"{col}_clean"] = problems[col].apply(
            lambda x: clean_html(str(x)) if pd.notna(x) else ""
        )

# Show samples
print("\nSample cleaned problems:")
for _, row in problems.head(5).iterrows():
    text = row["problem_text_clean"][:200]
    ans_type = row.get("Answer Type", "?")
    print(f"  ID {row['problem_id']} [{ans_type}]: {text}...")

# ============================================================
# Step 4: Aggregate skills per problem
# ============================================================
print("\nAggregating skills per problem...")
skills_agg = skills.groupby("problem_id").agg(
    skill_ids=("skill_id", lambda x: ";".join(str(i) for i in x)),
    skill_names=("node_name", lambda x: " | ".join(str(n) for n in x)),
    n_skills=("skill_id", "count"),
).reset_index()

# ============================================================
# Step 5: Compute item statistics
# ============================================================
print("Computing item-level statistics...")
item_stats = interactions.groupby("problem_id").agg(
    n_responses=("discrete_score", "count"),
    n_correct=("discrete_score", "sum"),
    p_value=("discrete_score", "mean"),
    mean_hints=("hint_count", "mean"),
    pct_saw_answer=("saw_answer", "mean"),
).reset_index()

# ============================================================
# Step 6: Compute student statistics
# ============================================================
print("Computing student-level statistics...")
student_stats = interactions.groupby("user_xid").agg(
    n_problems=("problem_id", "count"),
    n_unique_problems=("problem_id", "nunique"),
    n_correct=("discrete_score", "sum"),
    accuracy=("discrete_score", "mean"),
    total_hints=("hint_count", "sum"),
    pct_saw_answer=("saw_answer", "mean"),
).reset_index()

# ============================================================
# Step 7: Create merged problem table
# ============================================================
print("Creating master problem table...")
problem_master = problems.merge(item_stats, on="problem_id", how="left")
problem_master = problem_master.merge(skills_agg, on="problem_id", how="left")

# ============================================================
# Step 8: Save everything
# ============================================================
print("\nSaving processed files...")

# Problems with clean text + stats + skills
problem_master.to_parquet(os.path.join(OUT, "problems_master.parquet"), index=False)
problem_master.to_csv(os.path.join(OUT, "problems_master.csv"), index=False)

# Interactions (full)
interactions.to_parquet(os.path.join(OUT, "interactions.parquet"), index=False)

# Student stats
student_stats.to_parquet(os.path.join(OUT, "student_stats.parquet"), index=False)

# Item stats
item_stats.to_parquet(os.path.join(OUT, "item_stats.parquet"), index=False)

# Skills
skills.to_parquet(os.path.join(OUT, "skills.parquet"), index=False)

# Summary
summary = {
    "dataset": "FoundationalASSIST (processed)",
    "n_interactions": len(interactions),
    "n_problems": len(problems),
    "n_students": interactions["user_xid"].nunique(),
    "n_skills": skills["skill_id"].nunique(),
    "accuracy": float(interactions["discrete_score"].mean()),
    "answer_types": problems["Answer Type"].value_counts().to_dict() if "Answer Type" in problems.columns else {},
    "has_question_text": True,
    "has_distractor_text": True,
    "has_selected_answer": True,
    "has_demographics": False,
    "dif_note": "No demographics — contact etrials@assistments.org for demographic linkage",
}
with open(os.path.join(OUT, "summary.json"), "w") as f:
    json.dump(summary, f, indent=2, default=str)

print(f"\n{'='*60}")
print("PROCESSING COMPLETE!")
print(f"{'='*60}")
print(f"Output: {os.path.abspath(OUT)}/")
for f in sorted(os.listdir(OUT)):
    size = os.path.getsize(os.path.join(OUT, f)) / 1e6
    print(f"  {f}: {size:.1f} MB")

print(f"""
KEY FILES:
  problems_master.parquet — 3,395 problems with:
    - problem_text_clean (readable question text)
    - Multiple Choice Options/Answers (cleaned)
    - p_value, n_responses (item difficulty stats)
    - skill_names (Common Core alignment)
    
  interactions.parquet — 1.7M student responses with:
    - user_xid, problem_id, discrete_score (0/1)
    - answer_text (what the student actually typed/selected)
    - hint_count, saw_answer
    
  student_stats.parquet — per-student summary

NOTE: No demographic grouping variables available.
For DIF, you would need to:
  1. Contact etrials@assistments.org for demographic linkage, OR
  2. Create proficiency-based groups (e.g., top/bottom quartile) as proxy
""")