# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## Repository Overview

This is a Hugging Face dataset repository containing Canadian Parliamentary expenditure data. The dataset tracks spending by Members of Parliament from 2021 Q2 to 2025 Q4.

## Key Commands

### Dataset Management
```bash
# Activate the virtual environment (contains HF CLI)
source venv-hf/bin/activate

# Upload dataset to Hugging Face
hf upload irf23/canadian-parliamentary-expenditures . --repo-type dataset

# Upload only specific files
hf upload irf23/canadian-parliamentary-expenditures README.md --repo-type dataset

# Download latest files from HF (overwrites local files)
hf download irf23/canadian-parliamentary-expenditures --repo-type dataset --local-dir .

# Or use git to pull latest changes
git pull origin main
```

### Comparing Local vs Remote Changes
```bash
# See what's different between local and remote
git fetch origin
git diff HEAD origin/main

# See just file names that differ
git diff --name-only HEAD origin/main

# Download to a temporary directory to compare
hf download irf23/canadian-parliamentary-expenditures --repo-type dataset --local-dir /tmp/hf-compare
diff -r . /tmp/hf-compare

# Use git status to see local changes
git status
```

### Merging Strategies
```bash
# Option 1: Git merge (handles conflicts)
git fetch origin
git merge origin/main
# Resolve any conflicts, then commit

# Option 2: Download to temp, manually merge, then upload
hf download irf23/canadian-parliamentary-expenditures --repo-type dataset --local-dir /tmp/hf-latest
# Manually compare and merge files
# Then upload your merged version
hf upload irf23/canadian-parliamentary-expenditures . --repo-type dataset

# Option 3: Force push your version (overwrites remote)
hf upload irf23/canadian-parliamentary-expenditures . --repo-type dataset --commit-message "Force update with local changes"
```

### Data Validation
```bash
# Count expenditure records using pandas
python -c "import pandas as pd; df = pd.read_parquet('expenditures.parquet'); print(f'Total records: {len(df)}')"

# Count member records
python -c "import pandas as pd; df = pd.read_parquet('members.parquet'); print(f'Total members: {len(df)}')"

# Validate Parquet files
python -c "import pandas as pd; import glob; [pd.read_parquet(f) for f in glob.glob('*.parquet')]"

# Check analytics summary
jq . analytics.json
```

## Repository Structure

### Core Data Files
- `data/train/`: Training split containing:
  - `expenditures-{2021,2022,2023,2024}-*.parquet`: Quarterly files from 2021-2024
  - `members.parquet`: 450 parliament member records
- `data/test/`: Test split containing:
  - `expenditures-2025-*.parquet`: All 2025 quarterly files
  - `members.parquet`: 450 parliament member records (same as train)
- `analytics.json`: Analytics and summary statistics
- `metadata.json`: Dataset metadata with record counts
- `dataset_dict.json`: Defines dataset splits as ["expenditures", "members"]

### Visualization Apps
- `app.py`: Main Gradio app for dataset visualization
- `app_controlled.py`, `app_enhanced.py`, `app_free_tier.py`: Alternative versions with different features
- `space_files/`: Contains files for the separate Hugging Face Space

## Important Notes

1. **Dataset Updates**: When updating data files, always update `metadata.json` with new record counts
2. **README Configuration**: The README.md YAML frontmatter must not contain a `configs` section - this causes DuckDB errors
3. **Git Structure**: This repo uses a detached git history from the HF remote - use `--force` or `--force-with-lease` when pushing
4. **Virtual Environment**: The `venv-hf/` contains the Hugging Face CLI - always activate it before using HF commands
5. **Ignore Files**: The `.huggingfaceignore` excludes development files, virtual environments, and Space-specific files
6. **NEVER PUSH WITHOUT CONFIRMATION**: Always ask for user confirmation before pushing any changes to HuggingFace or executing git push commands

## Common Issues

### DuckDB Error on Dataset Page
- Caused by incorrect `configs` section in README.md
- Solution: Remove configs section, use proper `dataset_info` instead
- With Parquet files, ensure configs points to correct file patterns

### Push Conflicts
- The remote has different data structure (train/validation/test splits)
- Solution: Use `git push --force-with-lease` or HF CLI directly

### Authentication

#### HF CLI Login (Recommended)
```bash
# Login to HF (also configures git)
source venv-hf/bin/activate
hf auth login
# Or with token directly
hf auth login --token YOUR_HF_TOKEN
```

#### Git Credential Setup
```bash
# Option 1: Let HF CLI configure git (happens automatically with hf auth login)
hf auth login

# Option 2: Manually set git credentials with HF token
git config --global credential.helper store
git remote set-url origin https://YOUR_USERNAME:YOUR_HF_TOKEN@huggingface.co/datasets/irf23/canadian-parliamentary-expenditures

# Option 3: Use git credential helper (macOS)
git config --global credential.helper osxkeychain
# Git will prompt for username/password on first push
# Username: your HF username
# Password: your HF access token (not your password)

# Check current remote URL
git remote -v
```

#### Getting Your HF Token
1. Go to https://huggingface.co/settings/tokens
2. Create a new token with 'write' permissions
3. Copy the token (starts with 'hf_...')