Interactive Data Query Explorer

Query the OpenStreetMap Datasets in your Browser

This works by using DuckDB WASM to run SQL queries with data (.parquet files) after downloading them from Hugging Face. Keep in mind that these parquet file can be many MBs big. Visit dataset repository on Hugging Face to see all available data and file sizes.
Use the example queries to get started or enter your own query.
You can use an LLM (like ChatGPT, Mistral, etc.) to help generate custom SQL queries for this data. Paste the dataset information into your conversation to make it easier to generate queries.
OpenStreetMap Dataset Information for Custom Query Generation

This data is available for querying using DuckDB WASM in the browser. All data is stored as parquet files on Hugging Face at: https://huggingface.co/datasets/piebro/osm-data

IMPORTANT FILE SIZE NOTES:
- changeset_data: Each monthly parquet file is approximately 20 MB
- changeset_comments_data: Total dataset is 70-150 MB across all files
- notes_data: Total dataset is 30-36 MB across all files
- notes_comments_data: Total dataset is 40-60 MB across all files

Keep these file sizes in mind when writing queries - downloading large files may take time in the browser.

AVAILABLE DATASETS:

1. CHANGESET DATA
Location: https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=YYYY/month=M/data_0.parquet
Partitioned by: year and month

Key Columns:
- changeset_id: ID of the changeset
- edit_count: Number of edits in the changeset
- user_name: OSM contributor username
- year, month: Time partitioning columns
- created_by: Normalized editing software name (e.g., "iD", "JOSM", "StreetComplete")
- device_type: Classification (desktop_editor, mobile_editor, tool, other)
- bot: Boolean indicating if changeset was made by a bot
- mid_pos_x, mid_pos_y: Discretized coordinates (0-360, 0-180)
- imagery_used: Array of imagery sources
- hashtags: Array of hashtags from the changeset
- source: Array of data sources used
- mobile_os: Mobile OS (Android, iOS, or NULL)
- streetcomplete_quest: StreetComplete quest type (if applicable)
- organised_team: Team/corporation affiliation
- for_profit: Boolean for for-profit organisations

2. CHANGESET COMMENTS DATA
Location: https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_comments_data/part-0.parquet (multiple parts)

Key Columns:
- changeset_id: ID of changeset (join key with changeset_data)
- date: Timestamp of comment (UTC)
- user_name: Username of commenter
- text: Comment content

3. NOTES DATA
Location: https://huggingface.co/datasets/piebro/osm-data/resolve/main/notes_data/part-0.parquet (multiple parts)

Key Columns:
- note_id: Unique note identifier
- lat, lon: Latitude and longitude
- created_at: Creation timestamp (UTC)
- closed_at: Close timestamp (NULL if still open)
- mid_pos_x, mid_pos_y: Discretized coordinates (0-360, 0-180)

4. NOTES COMMENTS DATA
Location: https://huggingface.co/datasets/piebro/osm-data/resolve/main/notes_comments_data/part-0.parquet (multiple parts)

Key Columns:
- note_id: Note ID (join key with notes_data)
- action: Action type (opened, commented, closed, reopened)
- timestamp: Action timestamp (UTC)
- user_name: Username
- text: Comment content

DUCKDB QUERY PATTERNS:

Reading Parquet Files:
IMPORTANT: Wildcards ("*") cannot be used in HTTPS URLs. You must explicitly list each file path.

```sql
-- Single file
FROM read_parquet('https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=2025/month=1/data_0.parquet')

-- Multiple files (array) - each file must be listed explicitly
FROM read_parquet([
    'https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=2025/month=1/data_0.parquet',
    'https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=2025/month=2/data_0.parquet'
])

-- INVALID: This will NOT work with HTTPS URLs
-- FROM read_parquet('https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=*/month=*/*.parquet')
```

Working with Array Columns (imagery_used, hashtags, source):
```sql
-- Expand arrays with unnest
SELECT unnest(hashtags) as hashtag, edit_count
FROM read_parquet('...')
WHERE hashtags IS NOT NULL
```

Common Aggregations:
```sql
-- Cast to appropriate types for cleaner output
CAST(SUM(edit_count) as BIGINT) as "Total Edits"
CAST(COUNT(DISTINCT user_name) as BIGINT) as "Contributors"
COUNT(*) as "Total Records"
```

Time Grouping:
```sql
-- Format year-month for display
CONCAT(year, '-', LPAD(CAST(month as VARCHAR), 2, '0')) as months

-- Filter by date
WHERE year >= 2024
WHERE year = 2025 AND month >= 1
```

EXAMPLE QUERIES:

Top Software by Contributors (2025):
```sql
SELECT
    created_by as "Software",
    CAST(COUNT(DISTINCT user_name) as BIGINT) as "Contributors"
FROM read_parquet('https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=2025/month=1/data_0.parquet')
WHERE created_by IS NOT NULL
GROUP BY created_by
ORDER BY "Contributors" DESC
LIMIT 10
```

Monthly Edit Trend:
```sql
SELECT
    CONCAT(year, '-', LPAD(CAST(month as VARCHAR), 2, '0')) as "Month",
    CAST(SUM(edit_count) as BIGINT) as "Total Edits"
FROM read_parquet([
    'https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=2024/month=12/data_0.parquet',
    'https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=2025/month=1/data_0.parquet'
])
GROUP BY year, month
ORDER BY year, month
```

Bot vs Human Contributors:
```sql
SELECT
    bot,
    CAST(COUNT(DISTINCT user_name) as BIGINT) as "Contributors",
    CAST(SUM(edit_count) as BIGINT) as "Edits"
FROM read_parquet('https://huggingface.co/datasets/piebro/osm-data/resolve/main/changeset_data/year=2025/month=1/data_0.parquet')
GROUP BY bot
```

TIPS FOR QUERY GENERATION:
- Always use HTTPS URLs for parquet files
- CRITICAL: Wildcards ("*") do NOT work in HTTPS URLs - explicitly list each file path
- Remember that queries download data files - consider file sizes
- Use array syntax [...] for multiple files
- CAST aggregations to BIGINT or INTEGER for cleaner output
- Use meaningful column aliases with quotes: "Total Edits"
- Filter early (WHERE clauses) to reduce processing
- Use IS NOT NULL when filtering on nullable columns
- For recent data, focus on recent years/months to minimize download size