Philadelphia crime

Philadelphia Crime Dataset¶

Crime incidents from the Philadelphia Police Department. Part I crimes include violent offenses such as aggravated assault, rape, arson, among others. Part II crimes include simple assault, prostitution, gambling, fraud, and other non-violent offenses. Each record provides the date, time, and type of crime, allowing for both spatial and temporal analysis. For the benchmark, we rely on a subset of crime reports from 2023.

Run this notebook in Google Colab:

Remember to install the srai library before running the notebook:

%pip install srai[all]

In [1]:

Copied!





# plotting imports
import contextily as cx
import numpy as np
import seaborn as sns
from matplotlib import pyplot as plt
from matplotlib.patches import Patch

# dataset import
from srai.datasets import PhiladelphiaCrimeDataset
# plotting imports
import contextily as cx
import numpy as np
import seaborn as sns
from matplotlib import pyplot as plt
from matplotlib.patches import Patch

# dataset import
from srai.datasets import PhiladelphiaCrimeDataset

In [2]:

Copied!

philadelphia_crime = PhiladelphiaCrimeDataset()
philadelphia_crime = PhiladelphiaCrimeDataset()

In [3]:

Copied!

type(philadelphia_crime.train_gdf), type(philadelphia_crime.test_gdf)
type(philadelphia_crime.train_gdf), type(philadelphia_crime.test_gdf)

Out[3]:

(NoneType, NoneType)

Get data using .load() method -> a default version 'res_8'

In [4]:

Copied!

ds = philadelphia_crime.load()
ds.keys()
ds = philadelphia_crime.load()
ds.keys()

Downloading crime data for 2023...

Loading cached Parquet file for 2023...
Splitting into train-test subsets ...

Loading cached Parquet file for 2023...
Splitting into train-test subsets ...

Out[4]:

dict_keys(['train', 'test'])

In [5]:

Copied!

type(philadelphia_crime.train_gdf), type(philadelphia_crime.test_gdf)
type(philadelphia_crime.train_gdf), type(philadelphia_crime.test_gdf)

Out[5]:

(geopandas.geodataframe.GeoDataFrame, geopandas.geodataframe.GeoDataFrame)

In [6]:

Copied!

resolution = philadelphia_crime.resolution
resolution
resolution = philadelphia_crime.resolution
resolution

Out[6]:

In [7]:

Copied!

gdf_train, gdf_test = ds["train"], ds["test"]
gdf_train, gdf_test = ds["train"], ds["test"]

In [8]:

Copied!

gdf_train.head()
gdf_train.head()

Out[8]:

	objectid	dc_dist	psa	dispatch_date_time	dispatch_date	dispatch_time	hour	dc_key	location_block	ucr_general	text_general_code	point_x	point_y	geometry
411	116	01	1	2023-03-11 18:31:00	2023-03-11	13:31:00	13.0	2.023010e+11	2400 BLOCK S 28TH ST	600	Theft from Vehicle	-75.193618	39.922350	POINT (-75.19362 39.92235)
412	119	08	2	2023-03-11 22:13:00	2023-03-11	17:13:00	17.0	2.023080e+11	9800 BLOCK Roosevelt Blvd	600	Thefts	-75.015070	40.094525	POINT (-75.01507 40.09452)
413	96	15	1	2023-03-11 12:42:00	2023-03-11	07:42:00	7.0	2.023150e+11	4700 BLOCK GRISCOM ST	600	Thefts	-75.083953	40.017896	POINT (-75.08395 40.0179)
414	99	14	1	2023-03-12 00:54:00	2023-03-11	19:54:00	19.0	2.023140e+11	5500 BLOCK BLOYD ST	300	Robbery No Firearm	-75.161898	40.044952	POINT (-75.1619 40.04495)
416	102	25	3	2023-03-11 07:03:00	2023-03-11	02:03:00	2.0	2.023250e+11	200 BLOCK W ONTARIO ST	400	Aggravated Assault Firearm	-75.133172	40.002221	POINT (-75.13317 40.00222)

Getting the h3 with target values

In [9]:

Copied!

train_h3, _, test_h3 = philadelphia_crime.get_h3_with_labels()
train_h3, _, test_h3 = philadelphia_crime.get_h3_with_labels()

In [10]:

Copied!





fig, axes = plt.subplots(
    2, 1, sharex=False, sharey=False, figsize=(12, 16), height_ratios=[4, 1]
)

train_h3.plot(
    color="orange",
    markersize=0.1,
    ax=axes[0],
    label="train",
    alpha=np.minimum(np.power(train_h3[philadelphia_crime.target] + 0.4, 2), 1),
)
test_h3.plot(
    color="royalblue",
    markersize=0.1,
    ax=axes[0],
    label="test",
    alpha=np.minimum(np.power(test_h3[philadelphia_crime.target] + 0.4, 2), 1),
)

cx.add_basemap(axes[0], source=cx.providers.CartoDB.PositronNoLabels, crs=4326, zoom=12)
axes[0].set_title("Philadelphia crime data aggregated to H3 cells")
axes[0].legend(
    handles=[Patch(facecolor="orange"), Patch(facecolor="royalblue")],
    labels=["Train", "Test"],
)
axes[0].set_axis_off()

sns.kdeplot(
    x=train_h3[philadelphia_crime.target],
    label="train",
    color="orange",
    ax=axes[1],
    fill=False,
    cut=0,
)
sns.kdeplot(
    x=test_h3[philadelphia_crime.target],
    label="test",
    color="royalblue",
    ax=axes[1],
    fill=False,
    cut=0,
)
axes[1].set_title("Philadelphia crime data - target distribution")
axes[1].legend()

fig.tight_layout()

plt.show()
fig, axes = plt.subplots(
    2, 1, sharex=False, sharey=False, figsize=(12, 16), height_ratios=[4, 1]
)

train_h3.plot(
    color="orange",
    markersize=0.1,
    ax=axes[0],
    label="train",
    alpha=np.minimum(np.power(train_h3[philadelphia_crime.target] + 0.4, 2), 1),
)
test_h3.plot(
    color="royalblue",
    markersize=0.1,
    ax=axes[0],
    label="test",
    alpha=np.minimum(np.power(test_h3[philadelphia_crime.target] + 0.4, 2), 1),
)

cx.add_basemap(axes[0], source=cx.providers.CartoDB.PositronNoLabels, crs=4326, zoom=12)
axes[0].set_title("Philadelphia crime data aggregated to H3 cells")
axes[0].legend(
    handles=[Patch(facecolor="orange"), Patch(facecolor="royalblue")],
    labels=["Train", "Test"],
)
axes[0].set_axis_off()

sns.kdeplot(
    x=train_h3[philadelphia_crime.target],
    label="train",
    color="orange",
    ax=axes[1],
    fill=False,
    cut=0,
)
sns.kdeplot(
    x=test_h3[philadelphia_crime.target],
    label="test",
    color="royalblue",
    ax=axes[1],
    fill=False,
    cut=0,
)
axes[1].set_title("Philadelphia crime data - target distribution")
axes[1].legend()

fig.tight_layout()

plt.show()

No description has been provided for this image

Get data from 2013 year.

In [11]:

Copied!

ds = philadelphia_crime.load(version="2013")
ds.keys()
ds = philadelphia_crime.load(version="2013")
ds.keys()

Downloading crime data for 2013...

Loading cached Parquet file for 2013...

Out[11]:

dict_keys(['train'])

In [12]:

Copied!

type(philadelphia_crime.train_gdf), type(philadelphia_crime.test_gdf)
type(philadelphia_crime.train_gdf), type(philadelphia_crime.test_gdf)

Out[12]:

(geopandas.geodataframe.GeoDataFrame, NoneType)

In [13]:

Copied!

ds["train"].head()
ds["train"].head()

Out[13]:

	objectid	dc_dist	psa	dispatch_date_time	dispatch_date	dispatch_time	hour	dc_key	location_block	ucr_general	text_general_code	point_x	point_y	geometry
0	32951063	12	2	2013-04-24 04:00:00	2013-04-24	00:00:00	8	2.013120e+11	5900 BLOCK SPRINGFIELD AVE	100	Homicide - Criminal	-75.231532	39.935160	POINT (-75.23153 39.93516)
2	32951104	25	1	2013-04-06 04:00:00	2013-04-06	00:00:00	23	2.013250e+11	700 BLOCK W VENANGO STREET	100	Homicide - Criminal	-75.140959	40.006263	POINT (-75.14096 40.00626)
12	32951276	24	1	2013-12-01 05:00:00	2013-12-01	00:00:00	22	2.013241e+11	900 BLOCK E RUSSELL ST	100	Homicide - Criminal	-75.112436	40.000069	POINT (-75.11244 40.00007)
13	32951308	19	1	2013-01-03 05:00:00	2013-01-03	00:00:00	23	2.013190e+11	6000 BLOCK HADDINGTON LANE	100	Homicide - Criminal	-75.240554	39.977365	POINT (-75.24055 39.97736)
14	32951338	24	2	2013-11-17 05:00:00	2013-11-17	00:00:00	17	2.013241e+11	2900 BLOCK N KIP ST	100	Homicide - Criminal	-75.127046	39.993583	POINT (-75.12705 39.99358)

Creating your own train - test split -> Bucket regression (works similarly for spatial regression)

In [14]:

Copied!

philadelphia_crime.target
philadelphia_crime.target

Out[14]:

'count'

In [15]:

Copied!

train, test = philadelphia_crime.train_test_split(
    test_size=0.2, resolution=8, n_bins=10, random_state=42
)
train, test = philadelphia_crime.train_test_split(
    test_size=0.2, resolution=8, n_bins=10, random_state=42
)

Summary of the split:

  Train: 403 H3 cells (141029 points)
  Test: 117 H3 cells (35685 points)

  Expected ratios: {'train': 0.8, 'validation': 0, 'test': 0.2}
  Actual ratios: {'train': 0.798, 'test': 0.202}
  Actual ratios difference: {'train': 0.002, 'test': -0.002}
   bucket  train_ratio  test_ratio  train_ratio_difference  \
0       0      0.81421     0.18579                -0.01421   
1       1      0.80993     0.19007                -0.00993   
2       2      0.80693     0.19307                -0.00693   
3       3      0.79468     0.20532                 0.00532   
4       4      0.79568     0.20432                 0.00432   
5       5      0.79242     0.20758                 0.00758   
6       6      0.79190     0.20810                 0.00810   
7       7      0.80483     0.19517                -0.00483   
8       8      0.79277     0.20723                 0.00723   
9       9      0.80070     0.19930                -0.00070   

   test_ratio_difference  train_points  test_points  
0                0.01421           149           34  
1                0.00993           669          157  
2                0.00693          1793          429  
3               -0.00532          3259          842  
4               -0.00432          5156         1324  
5               -0.00758          8532         2235  
6               -0.00810         13380         3516  
7                0.00483         21876         5305  
8               -0.00723         30715         8029  
9                0.00070         55500        13814  
Created new train_gdf and test_gdf. Train len: 141029,test len: 35685

In [16]:

Copied!

type(philadelphia_crime.train_gdf), type(philadelphia_crime.test_gdf)
type(philadelphia_crime.train_gdf), type(philadelphia_crime.test_gdf)

Out[16]:

(geopandas.geodataframe.GeoDataFrame, geopandas.geodataframe.GeoDataFrame)

In [17]:

Copied!

philadelphia_crime.resolution
philadelphia_crime.resolution

Out[17]:

In [18]:

Copied!

train.head()
train.head()

Out[18]:

	objectid	dc_dist	psa	dispatch_date_time	dispatch_date	dispatch_time	hour	dc_key	location_block	ucr_general	text_general_code	point_x	point_y	geometry
2	32951104	25	1	2013-04-06 04:00:00	2013-04-06	00:00:00	23	2.013250e+11	700 BLOCK W VENANGO STREET	100	Homicide - Criminal	-75.140959	40.006263	POINT (-75.14096 40.00626)
12	32951276	24	1	2013-12-01 05:00:00	2013-12-01	00:00:00	22	2.013241e+11	900 BLOCK E RUSSELL ST	100	Homicide - Criminal	-75.112436	40.000069	POINT (-75.11244 40.00007)
13	32951308	19	1	2013-01-03 05:00:00	2013-01-03	00:00:00	23	2.013190e+11	6000 BLOCK HADDINGTON LANE	100	Homicide - Criminal	-75.240554	39.977365	POINT (-75.24055 39.97736)
14	32951338	24	2	2013-11-17 05:00:00	2013-11-17	00:00:00	17	2.013241e+11	2900 BLOCK N KIP ST	100	Homicide - Criminal	-75.127046	39.993583	POINT (-75.12705 39.99358)
15	32951344	39	3	2013-07-14 04:00:00	2013-07-14	00:00:00	1	2.013390e+11	2700 BLOCK N 27TH ST	100	Homicide - Criminal	-75.175501	39.996899	POINT (-75.1755 39.9969)

In [19]:

Copied!

test.head()
test.head()

Out[19]:

	objectid	dc_dist	psa	dispatch_date_time	dispatch_date	dispatch_time	hour	dc_key	location_block	ucr_general	text_general_code	point_x	point_y	geometry
0	32951063	12	2	2013-04-24 04:00:00	2013-04-24	00:00:00	8	2.013120e+11	5900 BLOCK SPRINGFIELD AVE	100	Homicide - Criminal	-75.231532	39.935160	POINT (-75.23153 39.93516)
18	32951355	17	2	2013-07-14 04:00:00	2013-07-14	00:00:00	21	2.013170e+11	2700 BLOCK REED ST	100	Homicide - Criminal	-75.189423	39.935696	POINT (-75.18942 39.9357)
28	32951596	24	1	2013-07-26 04:00:00	2013-07-26	00:00:00	1	2.013241e+11	2000 BLOCK PICKWICK ST	100	Homicide - Criminal	-75.099995	39.998954	POINT (-75.09999 39.99895)
36	32951761	16	1	2013-12-22 05:00:00	2013-12-22	00:00:00	21	2.013160e+11	3800 BLOCK RENO ST	100	Homicide - Criminal	-75.199156	39.968398	POINT (-75.19916 39.9684)
38	32951708	12	3	2013-10-16 04:00:00	2013-10-16	00:00:00	11	2.013121e+11	2400 BLOCK S 63RD ST	100	Homicide - Criminal	-75.228979	39.925023	POINT (-75.22898 39.92502)