partridge 1.1.1

Partridge is a python library for working with GTFS feeds using pandas DataFrames.

Partridge

Partridge is a Python 3.6+ library for working with GTFS feeds using pandas DataFrames.

Partridge is heavily influenced by our experience at Remix analyzing and debugging every GTFS feed we could find.

At the core of Partridge is a dependency graph rooted at trips.txt. Disconnected data is pruned away according to this graph when reading the contents of a feed.

Feeds can also be filtered to create a view specific to your needs. It’s most common to filter a feed down to specific dates (service_id) or routes (route_id), but any field can be filtered.

Philosphy

The design of Partridge is guided by the following principles:

As much as possible

• Favor speed

• Allow for extension

• Succeed lazily on expensive paths

• Fail eagerly on inexpensive paths

As little as possible

• Do anything other than efficiently read GTFS files into DataFrames

• Take an opinion on the GTFS spec

Installation

pip install partridge

GeoPandas support

pip install partridge[full]

Usage

Setup

import partridge as ptg

inpath = 'path/to/caltrain-2017-07-24/'

Inspecting the calendar

The date with the most trips

date, service_ids = ptg.read_busiest_date(inpath)
#  datetime.date(2017, 7, 17), frozenset({'CT-17JUL-Combo-Weekday-01'})

The week with the most trips

service_ids_by_date = ptg.read_busiest_week(inpath)
#  {datetime.date(2017, 7, 17): frozenset({'CT-17JUL-Combo-Weekday-01'}),
#   datetime.date(2017, 7, 18): frozenset({'CT-17JUL-Combo-Weekday-01'}),
#   datetime.date(2017, 7, 19): frozenset({'CT-17JUL-Combo-Weekday-01'}),
#   datetime.date(2017, 7, 20): frozenset({'CT-17JUL-Combo-Weekday-01'}),
#   datetime.date(2017, 7, 21): frozenset({'CT-17JUL-Combo-Weekday-01'}),
#   datetime.date(2017, 7, 22): frozenset({'CT-17JUL-Caltrain-Saturday-03'}),
#   datetime.date(2017, 7, 23): frozenset({'CT-17JUL-Caltrain-Sunday-01'})}

Dates with active service

service_ids_by_date = ptg.read_service_ids_by_date(path)

date, service_ids = min(service_ids_by_date.items())
#  datetime.date(2017, 7, 15), frozenset({'CT-17JUL-Caltrain-Saturday-03'})

date, service_ids = max(service_ids_by_date.items())
#  datetime.date(2019, 7, 20), frozenset({'CT-17JUL-Caltrain-Saturday-03'})

Dates with identical service

dates_by_service_ids = ptg.read_dates_by_service_ids(inpath)

busiest_date, busiest_service = ptg.read_busiest_date(inpath)
dates = dates_by_service_ids[busiest_service]

min(dates), max(dates)
#  datetime.date(2017, 7, 17), datetime.date(2019, 7, 19)

Reading a feed

_date, service_ids = ptg.read_busiest_date(inpath)

view = {
    'trips.txt': {'service_id': service_ids},
    'stops.txt': {'stop_name': 'Gilroy Caltrain'},
}

feed = ptg.load_feed(path, view)

Read shapes and stops as GeoDataFrames

service_ids = ptg.read_busiest_date(inpath)[1]
view = {'trips.txt': {'service_id': service_ids}}

feed = ptg.load_geo_feed(path, view)

feed.shapes.head()
#       shape_id                                           geometry
#  0  cal_gil_sf  LINESTRING (-121.5661454200744 37.003512297983...
#  1  cal_sf_gil  LINESTRING (-122.3944115638733 37.776439059278...
#  2   cal_sf_sj  LINESTRING (-122.3944115638733 37.776439059278...
#  3  cal_sf_tam  LINESTRING (-122.3944115638733 37.776439059278...
#  4   cal_sj_sf  LINESTRING (-121.9031703472137 37.330157067882...

minlon, minlat, maxlon, maxlat = feed.stops.total_bounds
#  -122.412076, 37.003485, -121.566088, 37.77639

Extracting a new feed

outpath = 'gtfs-slim.zip'

view = {'trips.txt': {'service_id': ptg.read_busiest_date(inpath)[1]}}

ptg.extract_feed(inpath, outpath, view)
feed = ptg.load_feed(outpath)

assert service_ids == set(feed.trips.service_id)

Features

• Surprisingly fast :)

• Load only what you need into memory

• Built-in support for resolving service dates

• Easily extended to support fields and files outside the official spec (TODO: document this)

• Handle nested folders and bad data in zips

• Predictable type conversions

Thank You

I hope you find this library useful. If you have suggestions for improving Partridge, please open an issue on GitHub.

History

1.1.1 (2019-09-13)

• Improve file encoding sniffer, which was misidentifying some Finnish/emoji unicode. Thanks to @dyakovlev!

1.1.0 (2019-02-21)

• Add partridge.load_geo_feed for reading stops and shapes into GeoPandas GeoDataFrames.

1.0.0 (2018-12-18)

This release is a combination of major internal refactorings and some minor interface changes. Overall, you should expect your upgrade from pre-1.0 versions to be relatively painless. A big thank you to @genhernandez and @csb19815 for their valuable design feedback. If you still need Python 2 support, please continue using version 0.11.0.

Here is a list of interface changes:

• The class partridge.gtfs.feed has been renamed to partridge.gtfs.Feed.

• The public interface for instantiating feeds is partridge.load_feed. This function replaces the previously undocumented function partridge.get_filtered_feed.

• A new function has been added for identifying the busiest week in a feed: partridge.read_busiest_date

• The public function partridge.get_representative_feed has been removed in favor of using partridge.read_busiest_date directly.

• The public function partridge.writers.extract_feed is now available via the top level module: partridge.extract_feed.

Miscellaneous minor changes:

• Character encoding detection is now done by the cchardet package instead of chardet. cchardet is faster, but may not always return the same result as chardet.

• Zip files are unpacked into a temporary directory instead of reading directly from the zip. These temporary directories are cleaned up when the feed is garbage collected or when the process exits.

• The code base is now annotated with type hints and the build runs mypy to verify the types.

• DataFrames are cached in a dictionary instead of the functools.lru_cache decorator.

• The partridge.extract_feed function now writes files concurrently to improve performance.

0.11.0 (2018-08-01)

• Fix major performance issue related to encoding detection. Thank you to @cjer for reporting the issue and advising on a solution.

0.10.0 (2018-04-30)

• Improved handling of non-standard compliant file encodings

• Only require functools32 for Python < 3

• ptg.parsers.parse_date no longer accepts dates, only strings

0.9.0 (2018-03-24)

• Improves read time for large feeds by adding LRU caching to ptg.parsers.parse_time.

0.8.0 (2018-03-14)

• Gracefully handle completely empty files. This change unifies the behavior of reading from a CSV with a header only (no data rows) and a completely empty (zero bytes) file in the zip.

0.7.0 (2018-03-09)

• Fix handling of nested folders and zip containing nested folders.

• Add ptg.get_filtered_feed for multi-file filtering.

0.6.1 (2018-02-24)

• Fix bug in ptg.read_service_ids_by_date. Reported by @cjer in #27.

0.6.0 (2018-02-21)

• Published package no longer includes unnecessary fixtures to reduce the size.

• Naively write a feed object to a zip file with ptg.write_feed_dangerously.

• Read the earliest, busiest date and its service_id’s from a feed with ptg.read_busiest_date.

• Bug fix: Handle calendar.txt/calendar_dates.txt entries w/o applicable trips.

0.6.0.dev1 (2018-01-23)

• Add support for reading files from a folder. Thanks again @danielsclint!

0.5.0 (2017-12-22)

• Easily build a representative view of a zip with ptg.get_representative_feed. Inspired by peartree.

• Extract out GTFS zips by agency_id/route_id with ptg.extract_{agencies,routes}.

• Read arbitrary files from a zip with feed.get('myfile.txt').

• Remove service_ids_by_date, dates_by_service_ids, and trip_counts_by_date from the feed class. Instead use ptg.{read_service_ids_by_date,read_dates_by_service_ids,read_trip_counts_by_date}.

0.4.0 (2017-12-10)

• Add support for Python 2.7. Thanks @danielsclint!

0.3.0 (2017-10-12)

• Fix service date resolution for raw_feed. Previously raw_feed considered all days of the week from calendar.txt to be active regardless of 0/1 value.

0.2.0 (2017-09-30)

• Add missing edge from fare_rules.txt to routes.txt in default dependency graph.

0.1.0 (2017-09-23)

• First release on PyPI.