Home Location Inference

from cider.datastore import DataStore
from cider.home_location import HomeLocator

Load some CDR data, antennas data, and (optionally) spatial files, whose paths have been defined in config.yml. Define the geographic granularity for inference (options are ‘antenna_id’, ‘tower_id’ if a tower_id is provided in the antennas files, or any of the spatial files). Optionally load wealth prediction data to construct a poverty map and ground truth location data to evaluate accuracy.

# This path should point to your cider installation, where configs and data for this demo are located.
from pathlib import Path
cider_installation_directory = Path('../../cider')

datastore = DataStore(config_file_path_string= cider_installation_directory / 'configs' / 'config_quickstart.yml')
home_locator = HomeLocator(datastore=datastore, clean_folders=True)

Loading CDR...
Loading antennas...
Warning: 10 antennas missing location

Filter to a specific date range.

home_locator.ds.filter_dates('2020-01-05', '2020-02-01')

Try running the three different home location algorithms, specifying towers as the requested geographic level of aggregation.

Note

The three algorithms currently implemented are relatively simple; more sophisticate methods to infer users’ home locations do exist.

home_locator.get_home_locations(algo='count_transactions', geo='tower_id')
home_locator.get_home_locations(algo='count_days', geo='tower_id')
home_locator.get_home_locations(algo='count_modal_days', geo='tower_id')

	subscriber_id	tower_id	count_modal_days
0	JGPCbfDGes	t35	4
1	YtvkGlMWwe	t39	4
2	dYwshzRseD	t26	4
3	dsBHAdXrrk	t35	2
4	ygMEXUQDbn	t27	4
...	...	...	...
995	amzyXHglBx	t7	3
996	dJSvXqUVSY	t64	3
997	ptudOnjdRU	t35	5
998	uXZrufHOmE	t79	4
999	zZkqaZFAtz	t80	4

1000 rows × 3 columns

Generate a population map based on inferred home locations using the unique days algorithm. Use points at each of the tower locations.

home_locator.map(algo='count_days', geo='tower_id', kind='population', voronoi=False)

Evaluate the accuracy of the home location inference using the ground truth data. Generate maps and a table of precision and recall for each tower.

home_locator.accuracy(algo='count_days', geo='tower_id')
home_locator.map(algo = 'count_days', geo='tower_id', kind='recall', voronoi=False)

Observations with inferred home location: 1000 (1000 unique)
Observations with ground truth home location: 1000 (1000 unique)
Observations with both: 1000 (1000 unique)
Overall accuracy: 0.02

home_locator.accuracy_tables[('tower_id', 'count_days')].head()

	tower_id	recall	precision	overall_accuracy
67	t7	0.076923	0.166667	0.017
15	t22	0.250000	0.142857	0.017
33	t39	0.214286	0.090909	0.017
66	t69	0.100000	0.055556	0.017
37	t42	0.153846	0.055556	0.017

Use population density maps to identify which areas are over/under-represented in your sample.

home_locator.pop_comparison(algo='count_days', geo='tower_id')