I have a CSV file of around 230 million rows, which contains a latitude column, a longitude column, a Date stamp, and an ID column. It look like this (I added the header in this example to help explain, the file does not have a header):

ID,                               Date,                     Lon,     lat
46d4089a713082d85452a2af64571644, 2016-11-30T12:57:11.000Z, 53.4529, -2.287
46d4089a713082d85452a2af64571644, 2016-11-30T12:57:26.000Z, 53.4521, -2.2859
46d4089a713082d85452a2af64571644, 2016-11-30T12:57:59.000Z, 53.4522, -2.2878
46d4089a713082d85452a2af64571644, 2016-11-30T12:59:01.000Z, 53.4547, -2.284
a6af7b30dc3ffa0ee7ecea02a2981b7d, 2016-11-30T13:03:01.000Z, 53.4457, -2.2774
693316c8b95e9fb07207400414714180, 2016-11-30T21:18:16.913Z, 53.2887, -2.6687

I also have two ESRI polygon shapefiles I created. What I want is a list of all the IDs that have at least one coordinate in both of the shapefiles.

Since the file is so large and could not be loaded into memory, my approach was to break the CSV file into 117 CSVs of 2 million rows. I then planned on using geopandas to read the smaller CSVs as a GeodataFrame, and use sjoin to find all the waypoints in shapefile 1.

I would then take the ID column of the waypoints within the shapefile as a list, take all waypoints in the large csv with an ID in the list, and do the same again finding the points in shapefile 2.

I tried to use spatial indexing to speed up the process, as it is a large file

I tried to test this using the first CSV of 2 million rows, using this code:

import pandas as pd
import geopandas
from geopandas.tools import sjoin
from shapely.geometry import *

waypoints = pd.read_csv(r'largefile_1.csv',sep=',', names=['TripId','lat','lon'],usecols=[0,2,3])

waypoints['geometry'] = waypoints.apply(lambda x: Point((float(x.lon), float(x.lat))), axis=1)

point = geopandas.GeoDataFrame(waypoints, geometry='geometry')
polygon  = geopandas.GeoDataFrame.from_file(r'Shapefile1.shp')
point.crs = polygon.crs

spatial_index = point.sindex
possible_matches_index = list(spatial_index.intersection(polygon.bounds))
possible_matches = point.iloc[possible_matches_index]
precise_matches = possible_matches[possible_matches.intersects(polygon)]

Is there a more efficient way of doing this, in another program or with different plugins?

  • 1
    Instead of keeping an unwieldy format and breaking it into an unmanageable number of pieces, you should probably be looking to use a tool which can handle that much data. File geodatabase or PostgreSQL/PostGIS spring to mind. Unfortunately, it is GIS SE policy to ask one question per Question, so you need to choose between debugging the hard way or asking for a different way. I would point out that you don't need to put all the points in memory, just the two polygons.
    – Vince
    Feb 1, 2017 at 12:24
  • I have edited my question so as to only ask about alternative methods. Feb 1, 2017 at 12:59
  • I have used SQL in the past, but never PostgreSQL. How would I go about approaching this problem? Feb 1, 2017 at 12:59
  • I would use the usual process: Database design, data loading, relational manipulation, analysis, and result processing. This is not the sort of task that lends itself to the GIS SE "Focused question / Best answer" model.
    – Vince
    Feb 1, 2017 at 13:38

1 Answer 1


If you have access to a postgres database (if not I would download it and install the extension postgis)

Here is an example how I uses python and psycopg2:

import psycopg2
import psycopg2.extras

class DBException(Exception):
class DB():
    def __init__(self, parent_widget):
        self.conn = None
        self.dbhost = 'localhost'
        self.dbname = 'dbname'# Name your db 
        self.dbuser = 'postgres' #If you dont have set it something else
        self.dbpass = 'postgres' #what ever your password is

    def _connect(self):
        Connects to the database
        if self.conn is None:
                self.conn = psycopg2.connect(
            except psycopg2.OperationalError as e:
                raise DBException("Error connecting to database on '%s'. %s" %
                                  (self.dbhost, str(e)))

    def _close(self):
        Closes the connection to the database
        if self.conn is not None:
            self.conn = None

    def create_table(self, sql, tbl_name):
        :param sql: text string with the sql statement
        :param tbl_name: text string
        cur = self.conn.cursor(cursor_factory=psycopg2.extras.DictCursor)
        cur.execute("DROP TABLE IF EXISTS " + str(tbl_name))

    def insert_data(self, sql):
        :param sql: text string with the sql statement
        cur = self.conn.cursor(cursor_factory=psycopg2.extras.DictCursor)
DB = DB()
DB.create_table("create table large_table(id varchar(40) PRIMARY KEY, date_insterted timestamp, pos geometry(POINT, 4326))" )

with open('the_csv_file.csv', 'r') as f:
    all_lines = f.readlines() # U might have to do some buffer work here
    sql_row_count = 0
    sql = ""
    for row in all_lines:
        row = row.split(',')
        if sql_row_count > 10000:
            sql = ""
            sql_row_count = 0

        sql += "Insert into large_table id, date_insterted, pos Values(" + row[0] + ", " + row[1]+", ST_PointFromText('POINT(" + row[2] + " " + row[3]) + ")',4326); "

That is a start to get them all into a postgres database, then you can write questions in postgres like "select * from large_table where st_x(pos) > 45 and st_x <55"

Hope that helps!

  • What's the purpose of the sql string? Feb 2, 2017 at 10:02
  • The purpose is to insert the data in the csv file into a postgres database. With DB.insert_data(sql).
    – axel_ande
    Feb 2, 2017 at 16:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.