geot-cluster 0.9.20

Reasearch project on clustering clients from transaction history

Name of the module stands for geographical transactions clustering. This module is an implementation of the method, developed for the third course project in HSE University. It takes dataframe with clients transactions history of the specified format and returns list of clusters.

For the record, it was intended to be for public usage in this form, as it is a research project seeking to find a way to deal with the described problem

Installation

Run the following to install:

‘’’python

pip install geot_cluster’’’

Usage

Before using make sure, that your dataset corresponds with requirements. Csv file must contain the following columns in order to work correctly

• user_id : string type, example: “423156821”

• event_dt : string type, example: “20190312”

• event_time: string type, example: “2019-03-12 06:24:00.279”

• lattitude : float type, example: 49.862621

• longtitude: see lattitude

Workflow:

import pandas as pd
import numpy  as np
import markov_clustering as mc
import matplotlib.pyplot as plt
import math
import pytz
import folium
import os.path
import networkx as nx


from haversine       import haversine, Unit
from collections     import Counter
from datetime        import datetime
from timezonefinder  import TimezoneFinder
from IPython.display import clear_output

import geotrans_cluster

path = [path to file with data]
data, names = data_load(path)


%matplotlib notebook
base = [path to the folder, where to store libs with information about clients]

archivate = True
libs= True
graph_f = True
cluster_f = True


if(archivate):
    archivate_maps(data, names, levels=4)

if(libs):
    lib = graph_preparation(data, names, base)
    prob_lib = znakomstvo_by_lib(lib,data)

lib, prob_lib = load_libs(base = base)

if(graph_f):
    graph = graph_forming(lib, prob_lib, treshold=0.9)

if(cluster_f):
    result = mc.run_mcl(graph,pruning_threshold=0.7, inflation=2,expansion=2)
    clusters = mc.get_clusters(result)

    clust_0 = clusters_to_ids(lib=lib, prob_lib=prob_lib, clusters = clusters, number = 0)
    maps = get_cluster_maps(data = data, clust = clust_0)
    print("Number of clusters", len(clusters))

    plt.figure(figsize=(10,10))
    mc.drawing.draw_graph(result, clusters, edge_color="red",node_size=15,width = 1, with_labels=True, font_size = 8)