README.md

---
tags:
- name
- lang
- iata
- city
- coordinates
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: code
    dtype: string
  - name: name
    dtype: string
  - name: lang
    dtype: string
  - name: name_en
    dtype: string
  - name: name_ru
    dtype: string
  - name: name_zh
    dtype: string
  - name: names_en_alt
    list: string
  - name: names_ru_alt
    list: string
  - name: names_zh_alt
    list: string
  - name: names_other
    list: string
  - name: country
    dtype: string
  - name: region
    dtype: string
  - name: links
    list: string
  - name: lat
    dtype: float64
  - name: lon
    dtype: float64
  splits:
  - name: train
    num_bytes: 2072877
    num_examples: 9026
  download_size: 3416192
  dataset_size: 2072877
---

# IATA City Codes with Lang and Coordinates

IATA city codes with language selection and geographic coordinates.

## Fields

- **code** (string): IATA metro code (e.g., 'NYC', 'MOW', 'SVO')
- **name** (string): Primary name in selected language
- **country** (string): ISO 3166-1 alpha-2 country code
- **region** (string): ISO 3166-2 subdivision code (if applicable)
- **name_en** (string): English name
- **name_ru** (string): Russian name
- **name_zh** (string): Chinese name
- **lang** (string): Language of primary name field (en, ru, or zh)
  - Determined by country's language rule
- **lat** (float): Latitude coordinate (decimal degrees, -90 to 90)
- **lon** (float): Longitude coordinate (decimal degrees, -180 to 180)
- **names_en_alt** (list): Alternative English names
- **names_ru_alt** (list): Alternative Russian names
- **names_zh_alt** (list): Alternative Chinese names
- **names_other** (list): Names in other languages
- **links** (list): Reference links

## Language Rule

Languages are determined by country using hardcoded mapping (same as iso_3166-1_alpha-2).

## Geographic Coverage

- **Coordinates**: 99.3% coverage (9057/9123 entries)
- **Format**: WGS84 (EPSG:4326)
- **Precision**: Approximately city-level

## Usage

```python
from datasets import load_dataset

ds = load_dataset('cterdam/iata-metro', split='train')

# Get Moscow metro
moscow = ds.filter(lambda x: x['code'] == 'MOW')[0]
print(moscow)
# {
#   'code': 'MOW',
#   'name': 'Москва',
#   'region': 'RU',
#   'subdiv': 'RU-MOW',
#   'lang': 'ru',
#   'lat': 55.7558,
#   'lon': 37.6173,
# }

# Find metros in China
china_metros = ds.filter(lambda x: x['country'] == 'CN')
print(f"Found {len(china_metros)} city entries in China")

# Geographic queries
nearby = ds.filter(lambda x: abs(x['lat'] - 40.7) < 1 and abs(x['lon'] + 74.0) < 1)
```

## See Also

- `cterdam/iso_3166-1_alpha-2`: Countries/regions with lang
- `cterdam/iso_3166-2`: Subdivisions with lang


## Invariants

All entries in this dataset maintain the following invariants (verified by `test/test_invariants.py`):

### Name Fields
- **Required**: Every entry has a `name` field, plus all three of
  `name_en`, `name_ru`, `name_zh` (no nulls in any of the three)
- **Constraint**: `name` always equals one of `name_en`, `name_ru`, or
  `name_zh` (based on `lang`)
- **Example**: If `lang: ru`, then `name == name_ru`
- **Sourcing**: Names are sourced in priority order from
  Wikipedia langlinks, then Wikidata labels, then machine translation
  (used only as a last resort for obscure entries)

### Lang Field
- **Required**: Every entry has a `lang` field
- **Values**: One of `en`, `ru`, or `zh`
- **Determination**: Hardcoded by region (not inferred from text)
  - **Slavic regions** (RU, UA, BY, KZ, etc.) → `lang: ru`
  - **Sinophone regions** (CN, TW, HK, JP, etc.) → `lang: zh`
  - **All others** → `lang: en`

## Verify Invariants

Clone the dataset and run tests:

\`\`\`bash
git clone https://huggingface.co/datasets/cterdam/iata-metro
cd iata-metro

# Install dependencies
pip install datasets

# Run invariant tests
pytest test/test_invariants.py -v
\`\`\`

Expected output: All tests pass, verifying dataset integrity.