Geopandas

Geopandas is an open-source project to make working with geospatial data in python easier.

GeoPandas extends the datatypes used by pandas to allow spatial operations on geometric types.

Geometric operations are performed by shapely. Geopandas further depends on fiona for file access and matplotlib for plotting.

LLM applications (chat, QA) that utilize geospatial data are an interesting area for exploration.

! pip install sodapy
! pip install pandas
! pip install geopandas

import ast

import geopandas as gpd
import pandas as pd
from langchain.document_loaders import OpenCityDataLoader

Create a GeoPandas dataframe from Open City Data as an example input.

# Load Open City Data
dataset = "tmnf-yvry"  # San Francisco crime data
loader = OpenCityDataLoader(city_id="data.sfgov.org", dataset_id=dataset, limit=5000)
docs = loader.load()

# Convert list of dictionaries to DataFrame
df = pd.DataFrame([ast.literal_eval(d.page_content) for d in docs])

# Extract latitude and longitude
df["Latitude"] = df["location"].apply(lambda loc: loc["coordinates"][1])
df["Longitude"] = df["location"].apply(lambda loc: loc["coordinates"][0])

# Create geopandas DF
gdf = gpd.GeoDataFrame(
    df, geometry=gpd.points_from_xy(df.Longitude, df.Latitude), crs="EPSG:4326"
)

# Only keep valid longitudes and latitudes for San Francisco
gdf = gdf[
    (gdf["Longitude"] >= -123.173825)
    & (gdf["Longitude"] <= -122.281780)
    & (gdf["Latitude"] >= 37.623983)
    & (gdf["Latitude"] <= 37.929824)
]

Visualization of the sample of SF crime data.

import matplotlib.pyplot as plt

# Load San Francisco map data
sf = gpd.read_file("https://data.sfgov.org/resource/3psu-pn9h.geojson")

# Plot the San Francisco map and the points
fig, ax = plt.subplots(figsize=(10, 10))
sf.plot(ax=ax, color="white", edgecolor="black")
gdf.plot(ax=ax, color="red", markersize=5)
plt.show()

Load GeoPandas dataframe as a Document for downstream processing (embedding, chat, etc).

The geometry will be the default page_content columns, and all other columns are placed in metadata.

But, we can specify the page_content_column.

from langchain.document_loaders import GeoDataFrameLoader

loader = GeoDataFrameLoader(data_frame=gdf, page_content_column="geometry")
docs = loader.load()

docs[0]

Document(page_content='POINT (-122.420084075249 37.7083109744362)', metadata={'pdid': '4133422003074', 'incidntnum': '041334220', 'incident_code': '03074', 'category': 'ROBBERY', 'descript': 'ROBBERY, BODILY FORCE', 'dayofweek': 'Monday', 'date': '2004-11-22T00:00:00.000', 'time': '17:50', 'pddistrict': 'INGLESIDE', 'resolution': 'NONE', 'address': 'GENEVA AV / SANTOS ST', 'x': '-122.420084075249', 'y': '37.7083109744362', 'location': {'type': 'Point', 'coordinates': [-122.420084075249, 37.7083109744362]}, ':@computed_region_26cr_cadq': '9', ':@computed_region_rxqg_mtj9': '8', ':@computed_region_bh8s_q3mv': '309', ':@computed_region_6qbp_sg9q': nan, ':@computed_region_qgnn_b9vv': nan, ':@computed_region_ajp5_b2md': nan, ':@computed_region_yftq_j783': nan, ':@computed_region_p5aj_wyqh': nan, ':@computed_region_fyvs_ahh9': nan, ':@computed_region_6pnf_4xz7': nan, ':@computed_region_jwn9_ihcz': nan, ':@computed_region_9dfj_4gjx': nan, ':@computed_region_4isq_27mq': nan, ':@computed_region_pigm_ib2e': nan, ':@computed_region_9jxd_iqea': nan, ':@computed_region_6ezc_tdp2': nan, ':@computed_region_h4ep_8xdi': nan, ':@computed_region_n4xg_c4py': nan, ':@computed_region_fcz8_est8': nan, ':@computed_region_nqbw_i6c3': nan, ':@computed_region_2dwj_jsy4': nan, 'Latitude': 37.7083109744362, 'Longitude': -122.420084075249})