Minimalist GeoBases implementation:
- no dependencies
- compatible with Python 3.9+, CPython and PyPy
- one data source: opentraveldata
- one Python module for easier distribution on clusters (like Hadoop)
- faster load time (5x)
- tested with pytest and tox
>>> from neobase import NeoBase
>>> b = NeoBase()
>>> b.get('ORY', 'city_code_list')
['PAR']
>>> b.get('ORY', 'city_name_list')
['Paris']
>>> b.get('ORY', 'country_code')
'FR'
>>> b.distance('ORY', 'CDG')
34.87...
>>> b.get_location('ORY')
LatLng(lat=48.72..., lng=2.35...)Use the Python package:
pip install neobaseCheck out readthedocs for the API.
You can customize the source data when initializing:
with open("file.csv") as f:
N = NeoBase(f)Otherwise the loaded file will be the embedded one, unless the OPTD_POR_FILE environment variable is set. In that case, it will load from the path defined in that variable.
You can manually retrieve the latest data source yourself too, but you expose yourself to some breaking changes if they occur in the data.
from io import StringIO
from urllib.request import urlopen
from neobase import NeoBase, OPTD_POR_URL
data = urlopen(OPTD_POR_URL).read().decode('utf8')
N = NeoBase(StringIO(data))
N.get("PAR")The reference date of validity can be changed as well:
N = NeoBase(date="2000-01-01")
N.get("AIY") # was decommissioned in 2015By default, the reference date will be set to today, unless the OPTD_POR_DATE environment variable is set. In that case, it will use that value.
You can customize the behavior regarding duplicates: points sharing the same IATA code, like NCE as airport and NCE as city. By default everything is kept, but you can set it so that only the first point with an IATA code is kept:
N = NeoBase(duplicates=False)
len(N) # about 10,000 "only"Note that you can use the OPTD_POR_DUPLICATES environment variable to control this as well: set it to 0 to drop duplicates.
Finally, you can customize fields loaded by subclassing.
class SubNeoBase(NeoBase):
KEY = 0 # iata_code
# Those loaded fields are the default ones
FIELDS = (
("name", 6, None),
("lat", 8, None),
("lng", 9, None),
("page_rank", 12, lambda s: float(s) if s else None),
("country_code", 16, None),
("country_name", 18, None),
('continent_name', 19, None),
("timezone", 31, None),
("city_code_list", 36, lambda s: s.split(",")),
('city_name_list', 37, lambda s: s.split('=')),
('location_type', 41, None),
("currency", 46, None),
)
N = SubNeoBase()You can query the data using:
python -m neobase PAR NCEtoxThe geographical operations like N.find_near("ORY", 100) or N.find_closest_from("ORY") perform a full scan of the data, and are not optimized (remember that this library has no dependencies by design).
If you want a more efficient solution, you should use a spatial index like a BallTree, for example using scikit-learn:
import numpy as np
from sklearn.neighbors import BallTree
from neobase import NeoBase
N = NeoBase()
iata_codes = []
coords = []
for key in N:
lat, lon = N.get_location(key)
if lat is not None and lon is not None:
iata_codes.append(N.get(key, "iata_code"))
coords.append([np.radians(lat), np.radians(lon)])
coords = np.array(coords)
tree = BallTree(coords, metric="haversine")
def find_closest_with_balltree(coord):
point = np.radians(coord)
_, idx = tree.query([point], k=1)
iata_code = iata_codes[idx[0][0]]
return iata_code
paris = (48.8566, 2.3522)
print(find_closest_with_balltree(paris)) # <0.1ms
print(list(N.find_closest_from_location(paris))) # ~30ms
