Frequencies and Chaining in Python-Pandas

Python Tools & Languages Pandasposted by Steve Miller March 24, 2020 Steve Miller

This article discusses chaining in Python. A few years ago, in a Q&A session following a presentation I gave on data...

This article discusses chaining in Python. A few years ago, in a Q&A session following a presentation I gave on data analysis (DA) to a group of college recruits for my then consulting company, I was asked to name what I considered the most important analytic technique. Though a surprise to the audience, my answer, counts and frequencies, was a no brainer for me.

I’ve made a living doing data analysis over the years. My variant of DA is a direct descendent of the exploratory data analysis (EDA) movement that emerged in the academic world 50 years ago as a reaction to the hyper mathematization of statistical science. The EDA approach, as a counterweight, sought to front and center basic data examination using simply understood tools at the expense of complex models whose assumptions were often dubious.

While there’s no precise definition of EDA, one characterization I like is: “EDA is an approach to data analysis that postpones the usual assumptions about what kind of model the data follow with the more direct approach of allowing the data itself to reveal its underlying structure and model.” In other words, EDA digs into the data, simply examining variables alone and in tandem.

In my work, DA starts with an extensive look at data distribution, both within and between attributes of interest. Measures of central tendency, order statistics/quantiles, density functions, problem data and outlier/anomaly detection, missing values, correlations etc. are critical, as are visualizations that support these techniques, such as dotplots, barplots, histograms, stripplots, boxplots, violinplots, densityplots, scatterplots, heatplots, and binplots. Add a trellis or small multiple capability to view distributions and relationships “by” values of additional attributes.

For all this, however, the most important DA technique for my money is one that delivers the distribution of frequencies and counts, both uni and multi-dimensional. And this is for all data types, including character, category, and numeric.

I’ve spent a fair amount of time and effort building frequency tabulation tools in R, Python-Pandas, and Julia. What follows are illustrations of multi-dimensional frequencies in Python-Pandas using Python’s syntax for “chaining” methods in sequence to accomplish multi-step tasks. By chaining is meant a multi-step pipeline that’s issued as one statement. Pandas directly supports only single attribute frequencies with its methods, but I show how the Pandas groupby().size() chain can be a building block of a generic multi-dimensional frequency’s capability.

The Python-Pandas code presented below starts with the simple uni-dimensional frequency case, then builds toward a more generic solution using chained Python-Pandas methods. In the end, a proof-of-concept frequencies function is introduced. To exercise the code, I use the trusty Chicago Crime 2001-present data set, which “reflects reported incidents of crime (with the exception of murders where data exists for each victim) that occurred in the City of Chicago from 2001 to present, minus the most recent seven days.” The daily data are available following a separate process that writes a portable “feather” file from the downloaded CSV and subsequently available to Python-Pandas, R and Julia. Alas, there are in excess of 7M records (1+ GB RAM) pertaining to the what, where, and when of Chicago crime over 20 years. Most analyses with this data involve counts — frequencies — among one or more dimensional attribute axes.

The technology used is Wintel 10 with 128 GB RAM, along with JupyterLab 1.2.4 and Python 3.7.5, plus foundation libraries Pandas 0.25.3 and Numpy 1.16.4. Feather 0.4.0 handles efficient disk storage of the dataframes.

[Related article: Frequencies and Chaining in Python-Pandas]

Import personal functions. flatten, blanks, metadf, mkdowncast, and freqsdf are used here.

In [1]:

import sys

functdir = "c:/steve/jupyter/notebooks/functions"
sys.path.append(functdir)

from pythonfunctions import *

blanks(2)

Next, import pertinent Python libraries. More than are necessary for these exercises.

In [2]:

import warnings, os, numpy as np, pandas as pd, time, datetime, re, string, random, math, feather
from collections import Counter
from functools import reduce

warnings.filterwarnings('ignore')

blanks(2)

Read the embellished Chicago crime data from an R/Python portable feather file generated by an earlier running R script. 7M+ cases of 24 attributes. Compute year, month and day features from the datetime field. Detail the metadata. Pretty fast.

In [3]:

start = time.time()

wdir = "c:/bigdata/raw/chicago"
os.chdir(wdir)

ifile = "odsccrimeplus.feather"
chicagocrime = feather.read_dataframe(ifile).reset_index(drop=True)

chicagocrime.eval("""
year = date.dt.year
month = date.dt.month
day = date.dt.day
""",inplace=True)

metadf(chicagocrime)

end = time.time()
print(end - start)

blanks(2)

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 7092612 entries, 0 to 7092611
Data columns (total 27 columns):
id                     int32
casenumber             object
date                   datetime64[ns, UTC]
block                  object
iucr                   object
primarytype            object
description            object
locationdescription    object
arrest                 bool
domestic               bool
beat                   int32
district               float64
ward                   float64
communityarea          float64
fbicode                object
xcoordinate            float64
ycoordinate            float64
year                   int64
updatedon              datetime64[ns, UTC]
latitude               float64
longitude              float64
fbicodedesc            object
name                   object
population             float64
month                  int64
day                    int64
hour                   int32
dtypes: bool(2), datetime64[ns, UTC](2), float64(8), int32(3), int64(3), object(9)
memory usage: 1.3+ GB
None


8.357961177825928

“Downcast” float and integer attributes of chicagocrime to conserve RAM — from 1.3 GB to .9 GB.

In [4]:

start = time.time()

chicagocrime = mkdowncast(chicagocrime)

metadf(chicagocrime)
blanks(1)

end = time.time()
print(end - start)

blanks(2)

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 7092612 entries, 0 to 7092611
Data columns (total 27 columns):
id                     int32
casenumber             object
date                   datetime64[ns, UTC]
block                  object
iucr                   object
primarytype            object
description            object
locationdescription    object
arrest                 bool
domestic               bool
beat                   int16
district               float32
ward                   float32
communityarea          float32
fbicode                object
xcoordinate            float32
ycoordinate            float32
year                   int16
updatedon              datetime64[ns, UTC]
latitude               float32
longitude              float32
fbicodedesc            object
name                   object
population             float32
month                  int8
day                    int8
hour                   int8
dtypes: bool(2), datetime64[ns, UTC](2), float32(8), int16(2), int32(1), int8(3), object(9)
memory usage: 899.6+ MB
None




2.7455520629882812

Demo the Pandas value_counts method for 1-dimensional frequencies. By default, NA’s are excluded, so the sum of frequencies < #records.

In [5]:

freqs = pd.DataFrame(chicagocrime['name'].value_counts())
print(freqs.name.sum())
print(freqs)

blanks(1)

freqs = pd.DataFrame(chicagocrime['name'].value_counts(dropna=False))
print(freqs.name.sum())
print(freqs)

blanks(2)

6479024
                   name
Austin           410333
Near North Side  225363
South Shore      211725
Humboldt Park    204517
Near West Side   192609
...                 ...
Oakland           14323
Hegewisch         14008
Forest Glen       11628
Burnside           9689
Edison Park        6272

[77 rows x 1 columns]


7092612
                   name
NaN              613588
Austin           410333
Near North Side  225363
South Shore      211725
Humboldt Park    204517
...                 ...
Oakland           14323
Hegewisch         14008
Forest Glen       11628
Burnside           9689
Edison Park        6272

[78 rows x 1 columns]

From this point forward, the attributes of frequency interest will be delineated in the list “var”. Next consider Pandas groupby().size() chaining for a general frequencies’ solution. By default, a Pandas series is returned. Adding reset_index converts the series to a dataframe. Note that NA’s are ignored.

In [6]:

var = ['name']
freqs = chicagocrime[var].groupby(var).size()
print(freqs.sum(),"\n")
print(freqs)
blanks(1)
print(type(freqs))
blanks(1)

freqs = chicagocrime[var].groupby(var).size().reset_index()
print(freqs.iloc[:,1].sum(),"\n")
print(freqs)
blanks(1)
print(type(freqs))

blanks(2)

6479024 

name
Albany Park        57799
Archer Heights     23259
Armour Square      24116
Ashburn            59028
Auburn Gresham    183696
                   ...  
West Lawn          48320
West Pullman      106090
West Ridge         81138
West Town         190911
Woodlawn          105282
Length: 77, dtype: int64


<class 'pandas.core.series.Series'>


6479024 

              name       0
0      Albany Park   57799
1   Archer Heights   23259
2    Armour Square   24116
3          Ashburn   59028
4   Auburn Gresham  183696
..             ...     ...
72       West Lawn   48320
73    West Pullman  106090
74      West Ridge   81138
75       West Town  190911
76        Woodlawn  105282

[77 rows x 2 columns]


<class 'pandas.core.frame.DataFrame'>

What to do? — it’s critical to include NA’s in groupings! One solution: convert NA’s to a “special” not null value — in this example math.inf — before grouping, via the additional chaining step fillna. There are, of course, risks with this approach, if math.inf values appear naturally in the data. Below is a 2-way frequencies of name and district using this strategy. Note that the total count equals the number of records, i.e. includes NA’s converted to MISSING.

In [7]:

MISSING = math.inf

var = ['name','district']

freqs = chicagocrime[var].fillna(MISSING).groupby(var).size().reset_index()

print(freqs.iloc[:,2].sum())
metadf(freqs,data=True)
blanks(2)

7092612


<class 'pandas.core.frame.DataFrame'>
RangeIndex: 416 entries, 0 to 415
Data columns (total 3 columns):
name        416 non-null object
district    416 non-null float64
0           416 non-null int64
dtypes: float64(1), int64(1), object(1)
memory usage: 9.9+ KB
None


  name  district      0
0  inf       1.0  20007
1  inf       2.0  42898
2  inf       3.0  30632
3  inf       4.0  31962
4  inf       5.0  25770


          name  district       0
411  West Town       inf       4
412   Woodlawn       2.0    1072
413   Woodlawn       3.0  104206
414   Woodlawn       7.0       2
415   Woodlawn       inf       2

Add a step to convert the MISSING=math.inf back to NA’s after the grouping.

In [8]:

MISSING = math.inf

var = ['name','district']

freqs = chicagocrime[var].fillna(MISSING).groupby(var).size().reset_index().replace(MISSING,np.NaN)

print(freqs.iloc[:,2].sum(),"\n")
metadf(freqs,data=True)
blanks(2)

7092612 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 416 entries, 0 to 415
Data columns (total 3 columns):
name        393 non-null object
district    387 non-null float64
0           416 non-null int64
dtypes: float64(1), int64(1), object(1)
memory usage: 9.9+ KB
None


  name  district      0
0  NaN       1.0  20007
1  NaN       2.0  42898
2  NaN       3.0  30632
3  NaN       4.0  31962
4  NaN       5.0  25770


          name  district       0
411  West Town       NaN       4
412   Woodlawn       2.0    1072
413   Woodlawn       3.0  104206
414   Woodlawn       7.0       2
415   Woodlawn       NaN       2

Next, additional chaining steps to rename freqs attributes and sort by frequency. Also a last reset_index.

In [9]:

MISSING = math.inf

var = ['name','district']

freqs = chicagocrime[var].fillna(MISSING).groupby(var).size()\
.reset_index().replace(MISSING,np.NaN)\
.rename(columns={0:'frequency'}).sort_values(['frequency']+var, ascending=False)\
.reset_index(drop=True)

print(freqs.iloc[:,2].sum(),"\n")
metadf(freqs,data=True)
blanks(2)

7092612 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 416 entries, 0 to 415
Data columns (total 3 columns):
name         393 non-null object
district     387 non-null float64
frequency    416 non-null int64
dtypes: float64(1), int64(1), object(1)
memory usage: 9.9+ KB
None


              name  district  frequency
0           Austin      15.0     281985
1  Near North Side      18.0     225330
2   West Englewood       7.0     188959
3        Englewood       7.0     167079
4     Chicago Lawn       8.0     159647


               name  district  frequency
411         Ashburn      10.0          1
412         Ashburn       7.0          1
413  Archer Heights      31.0          1
414     Albany Park       8.0          1
415     Albany Park       NaN          1

Chain a final assign step to compute percent of total with a Python lambda function.

In [10]:

MISSING = math.inf

var = ['name','district']

freqs = chicagocrime[var].fillna(MISSING).groupby(var).size()\
.reset_index().replace(MISSING,np.NaN)\
.rename(columns={0:'frequency'}).sort_values(['frequency']+var, ascending=False)\
.assign(percent=lambda x: 100*(x.frequency/x.frequency.sum())).reset_index(drop=True)

print(freqs.iloc[:,2].sum(),"\n")
metadf(freqs,data=True)
blanks(2)

7092612 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 416 entries, 0 to 415
Data columns (total 4 columns):
name         393 non-null object
district     387 non-null float64
frequency    416 non-null int64
percent      416 non-null float64
dtypes: float64(2), int64(1), object(1)
memory usage: 13.1+ KB
None


              name  district  frequency   percent
0           Austin      15.0     281985  3.975757
1  Near North Side      18.0     225330  3.176968
2   West Englewood       7.0     188959  2.664167
3        Englewood       7.0     167079  2.355677
4     Chicago Lawn       8.0     159647  2.250891


               name  district  frequency   percent
411         Ashburn      10.0          1  0.000014
412         Ashburn       7.0          1  0.000014
413  Archer Heights      31.0          1  0.000014
414     Albany Park       8.0          1  0.000014
415     Albany Park       NaN          1  0.000014

Examine NA name and district in the latest freqs. Many more NA names.

In [11]:

print(freqs.loc[pd.isna(freqs.name)])
blanks(1)
print(freqs.loc[pd.isna(freqs.district)])

blanks(2)

    name  district  frequency   percent
48   NaN       2.0      42898  0.604827
54   NaN       8.0      39436  0.556015
61   NaN      11.0      35774  0.504384
62   NaN      25.0      35220  0.496573
66   NaN       7.0      33730  0.475565
68   NaN       6.0      32182  0.453740
70   NaN       4.0      31962  0.450638
71   NaN      12.0      31779  0.448058
73   NaN       9.0      30963  0.436553
74   NaN       3.0      30632  0.431886
75   NaN      19.0      29005  0.408947
77   NaN      18.0      28524  0.402165
79   NaN      14.0      28230  0.398020
86   NaN      15.0      25971  0.366170
87   NaN       5.0      25770  0.363336
93   NaN      10.0      24817  0.349899
106  NaN       1.0      20007  0.282082
107  NaN      24.0      19093  0.269196
108  NaN      22.0      19057  0.268688
109  NaN      17.0      18280  0.257733
110  NaN      16.0      17779  0.250669
121  NaN      20.0      12459  0.175662
161  NaN      31.0         20  0.000282


                       name  district  frequency   percent
185               West Town       NaN          4  0.000056
192         Near North Side       NaN          4  0.000056
201           Humboldt Park       NaN          3  0.000042
206                  Austin       NaN          3  0.000042
210                Woodlawn       NaN          2  0.000028
218         Washington Park       NaN          2  0.000028
219      Washington Heights       NaN          2  0.000028
222                The Loop       NaN          2  0.000028
226          South Lawndale       NaN          2  0.000028
236            Lincoln Park       NaN          2  0.000028
243               Englewood       NaN          2  0.000028
248            Chicago Lawn       NaN          2  0.000028
261              West Ridge       NaN          1  0.000014
265      West Garfield Park       NaN          1  0.000014
269          West Englewood       NaN          1  0.000014
289                Roseland       NaN          1  0.000014
302            Portage Park       NaN          1  0.000014
307                  O'Hare       NaN          1  0.000014
312            North Center       NaN          1  0.000014
318          Near West Side       NaN          1  0.000014
329         Lower West Side       NaN          1  0.000014
346               Lake View       NaN          1  0.000014
356                 Hermosa       NaN          1  0.000014
358  Greater Grand Crossing       NaN          1  0.000014
375                 Dunning       NaN          1  0.000014
397          Belmont Cragin       NaN          1  0.000014
402             Avalon Park       NaN          1  0.000014
409          Auburn Gresham       NaN          1  0.000014
415             Albany Park       NaN          1  0.000014

Define a simplistic frequencies function building on the chaining approach above. Does not handle category attributes, but is suitable here.

In [12]:

def freqsdf(df,var):
    """
    (df - pandas dataframe; var - list of dataframe columns)
    """
    
    MISSING = math.inf
    
    if len(var) != len(set(var)):
        print("duplicate columns\n")
        return(None)
        
    if not all(v in list(df.columns) for v in var):
        print("invalid column(s)")
        return(None)
        
    if len([str(v) for v in df[var].dtypes if str(v)=="category"]) > 0:
        print("invalid category column(s)")
        return(None)
  
    freqs = df[var].fillna(MISSING).groupby(var).size().reset_index().replace(MISSING,np.NaN)\
    .rename(columns={0:'frequency'}).sort_values(['frequency']+var, ascending=False)\
    .assign(percent=lambda x: 100*(x.frequency/x.frequency.sum())).reset_index(drop=True)
    
    return(freqs)

Quick first test with 2 attributes.

In [13]:

var = ['fbicode','fbicodedesc']
freqs = freqsdf(chicagocrime,var)

print(freqs.frequency.sum(),"\n")
metadf(freqs,data=True)

blanks(2)

7092612 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 26 entries, 0 to 25
Data columns (total 4 columns):
fbicode        26 non-null object
fbicodedesc    26 non-null object
frequency      26 non-null int64
percent        26 non-null float64
dtypes: float64(1), int64(1), object(2)
memory usage: 960.0+ bytes
None


  fbicode             fbicodedesc  frequency    percent
0      06                 Larceny    1500365  21.153913
1     08B          Simple Battery    1108831  15.633606
2      14               Vandalism     806114  11.365545
3      26  Misc Non-Index Offense     716041  10.095590
4      18              Drug Abuse     686916   9.684951


   fbicode                fbicodedesc  frequency   percent
21      09                      Arson      11535  0.162634
22     01A  Homicide 1st & 2nd Degree      10084  0.142176
23      13            Stolen Property       2839  0.040028
24      12               Embezzlement       1475  0.020796
25     01B   Involuntary Manslaughter         51  0.000719

Does it work with numeric variables? Yes, an 860,000+ record frequency table!

In [14]:

start = time.time()

var = ['latitude','longitude']
freqs = freqsdf(chicagocrime,var)

print(freqs.frequency.sum(),"\n")
metadf(freqs,data=True)
blanks(1)

end = time.time()
print(end - start)
blanks(2)

7092612 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 867302 entries, 0 to 867301
Data columns (total 4 columns):
latitude     867301 non-null float64
longitude    867301 non-null float64
frequency    867302 non-null int64
percent      867302 non-null float64
dtypes: float64(3), int64(1)
memory usage: 26.5 MB
None


    latitude  longitude  frequency   percent
0        NaN        NaN      68077  0.959830
1  41.976292 -87.905228      13518  0.190593
2  41.754593 -87.741531       9813  0.138355
3  41.883499 -87.627876       7655  0.107929
4  41.897896 -87.624100       4663  0.065744


         latitude  longitude  frequency   percent
867297  41.644604 -87.609550          1  0.000014
867298  41.644604 -87.610893          1  0.000014
867299  41.644604 -87.611778          1  0.000014
867300  41.644600 -87.597176          1  0.000014
867301  41.644588 -87.616081          1  0.000014


2.6000795364379883

Specify crime types by groupings of fbicodes to be used for filtering. hcde is homicide; vcde is violent crime; pcde is property crime; and icde is index crime.

In [15]:

hcde = ["01A"]

vcde = list(flatten([hcde,"01B","02","03","04A","04B"]))

pcde = ["05","06","07","09"]

icde = list(flatten([vcde,pcde]))

Compute/store a few booleans filtering crime type.

In [16]:

chicagocrime.eval("""
hbool = fbicode in @hcde
vbool = fbicode in @vcde
pbool = fbicode in @pcde
ibool = fbicode in @icde
""",inplace=True)

metadf(chicagocrime)

blanks(2)

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 7092612 entries, 0 to 7092611
Data columns (total 31 columns):
id                     int32
casenumber             object
date                   datetime64[ns, UTC]
block                  object
iucr                   object
primarytype            object
description            object
locationdescription    object
arrest                 bool
domestic               bool
beat                   int16
district               float32
ward                   float32
communityarea          float32
fbicode                object
xcoordinate            float32
ycoordinate            float32
year                   int16
updatedon              datetime64[ns, UTC]
latitude               float32
longitude              float32
fbicodedesc            object
name                   object
population             float32
month                  int8
day                    int8
hour                   int8
hbool                  bool
vbool                  bool
pbool                  bool
ibool                  bool
dtypes: bool(6), datetime64[ns, UTC](2), float32(8), int16(2), int32(1), int8(3), object(9)
memory usage: 926.7+ MB
None

3-way frequencies.

In [17]:

var = ['name','district','fbicodedesc']
freqs = freqsdf(chicagocrime,var)

print(freqs.frequency.sum(),"\n")
metadf(freqs,data=True)

blanks(2)

7092612 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4211 entries, 0 to 4210
Data columns (total 5 columns):
name           3655 non-null object
district       4165 non-null float64
fbicodedesc    4211 non-null object
frequency      4211 non-null int64
percent        4211 non-null float64
dtypes: float64(2), int64(1), object(2)
memory usage: 164.6+ KB
None


              name  district     fbicodedesc  frequency   percent
0  Near North Side      18.0         Larceny      94479  1.332076
1         The Loop       1.0         Larceny      82013  1.156316
2           Austin      15.0      Drug Abuse      65297  0.920634
3           Austin      15.0  Simple Battery      52223  0.736301
4        Lake View      19.0         Larceny      48218  0.679834


     name  district     fbicodedesc  frequency   percent
4206  NaN      31.0  Simple Assault          1  0.000014
4207  NaN      31.0      Drug Abuse          1  0.000014
4208  NaN      20.0    Embezzlement          1  0.000014
4209  NaN      10.0    Embezzlement          1  0.000014
4210  NaN       5.0    Embezzlement          1  0.000014

Same, filtering for homicides only.

In [18]:

var = ['name','district','fbicodedesc']
freqs = freqsdf(chicagocrime.query('fbicode in @hcde'),var)

print(freqs.frequency.sum(),"\n")
metadf(freqs,data=True)

blanks(2)

10084 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 140 entries, 0 to 139
Data columns (total 5 columns):
name           119 non-null object
district       140 non-null float64
fbicodedesc    140 non-null object
frequency      140 non-null int64
percent        140 non-null float64
dtypes: float64(2), int64(1), object(2)
memory usage: 5.6+ KB
None


             name  district                fbicodedesc  frequency   percent
0          Austin      15.0  Homicide 1st & 2nd Degree        600  5.950020
1  West Englewood       7.0  Homicide 1st & 2nd Degree        419  4.155097
2       Englewood       7.0  Homicide 1st & 2nd Degree        413  4.095597
3  North Lawndale      10.0  Homicide 1st & 2nd Degree        364  3.609679
4        New City       9.0  Homicide 1st & 2nd Degree        329  3.262594


              name  district                fbicodedesc  frequency   percent
135    Fuller Park       2.0  Homicide 1st & 2nd Degree          1  0.009917
136    Forest Glen      17.0  Homicide 1st & 2nd Degree          1  0.009917
137      Englewood       3.0  Homicide 1st & 2nd Degree          1  0.009917
138        Chatham       4.0  Homicide 1st & 2nd Degree          1  0.009917
139  Armour Square       2.0  Homicide 1st & 2nd Degree          1  0.009917

Homicides version II.

In [19]:

var = ['name','district','fbicodedesc']
freqs = freqsdf(chicagocrime.query('hbool'),var)

print(freqs.frequency.sum(),"\n")
metadf(freqs,data=True)

blanks(2)

10084 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 140 entries, 0 to 139
Data columns (total 5 columns):
name           119 non-null object
district       140 non-null float64
fbicodedesc    140 non-null object
frequency      140 non-null int64
percent        140 non-null float64
dtypes: float64(2), int64(1), object(2)
memory usage: 5.6+ KB
None


             name  district                fbicodedesc  frequency   percent
0          Austin      15.0  Homicide 1st & 2nd Degree        600  5.950020
1  West Englewood       7.0  Homicide 1st & 2nd Degree        419  4.155097
2       Englewood       7.0  Homicide 1st & 2nd Degree        413  4.095597
3  North Lawndale      10.0  Homicide 1st & 2nd Degree        364  3.609679
4        New City       9.0  Homicide 1st & 2nd Degree        329  3.262594


              name  district                fbicodedesc  frequency   percent
135    Fuller Park       2.0  Homicide 1st & 2nd Degree          1  0.009917
136    Forest Glen      17.0  Homicide 1st & 2nd Degree          1  0.009917
137      Englewood       3.0  Homicide 1st & 2nd Degree          1  0.009917
138        Chatham       4.0  Homicide 1st & 2nd Degree          1  0.009917
139  Armour Square       2.0  Homicide 1st & 2nd Degree          1  0.009917

More complicated filter.

In [20]:

var = ['name','district','fbicodedesc','year']
freqs = freqsdf(chicagocrime.query('vbool and 2015<year<2020'),var)

print(freqs.frequency.sum(),"\n")
metadf(freqs,data=True)

blanks(2)

106179 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 2264 entries, 0 to 2263
Data columns (total 6 columns):
name           2263 non-null object
district       2264 non-null float64
fbicodedesc    2264 non-null object
year           2264 non-null int64
frequency      2264 non-null int64
percent        2264 non-null float64
dtypes: float64(2), int64(2), object(2)
memory usage: 106.2+ KB
None


     name  district         fbicodedesc  year  frequency   percent
0  Austin      15.0             Robbery  2016        596  0.561316
1  Austin      15.0             Robbery  2017        589  0.554724
2  Austin      15.0  Aggravated Battery  2016        528  0.497273
3  Austin      15.0             Robbery  2018        522  0.491623
4  Austin      15.0  Aggravated Battery  2019        467  0.439823


                name  district                fbicodedesc  year  frequency  \
2259   Armour Square       1.0         Aggravated Assault  2017          1   
2260  Archer Heights       8.0  Homicide 1st & 2nd Degree  2018          1   
2261     Albany Park      17.0  Homicide 1st & 2nd Degree  2019          1   
2262     Albany Park      17.0  Homicide 1st & 2nd Degree  2017          1   
2263             NaN      12.0    Criminal Sexual Assault  2018          1   

       percent  
2259  0.000942  
2260  0.000942  
2261  0.000942  
2262  0.000942  
2263  0.000942

Create a season attribute in chicagocrime for subsequent frequencies. Drop it after use.

In [21]:

def mkseason(n):
    if 3<=n<=5: season = 'spring'
    elif 6<=n<=8: season = 'summer'
    elif 9<=n<=11: season = 'fall'
    elif n<=2 or n==12: season = 'winter'
    else: season = 'unknown'
    return(season)

chicagocrime['season'] = chicagocrime['month'].apply(mkseason)


var = ['name','district','fbicodedesc','season']
freqs = freqsdf(chicagocrime.query('vbool'),var)

print(freqs.frequency.sum(),"\n")
metadf(freqs,data=True)

chicagocrime.drop(['season'], axis=1)

blanks(2)

614873 



<class 'pandas.core.frame.DataFrame'>
RangeIndex: 3012 entries, 0 to 3011
Data columns (total 6 columns):
name           2581 non-null object
district       3004 non-null float64
fbicodedesc    3012 non-null object
season         3012 non-null object
frequency      3012 non-null int64
percent        3012 non-null float64
dtypes: float64(2), int64(1), object(3)
memory usage: 141.3+ KB
None


             name  district         fbicodedesc  season  frequency   percent
0          Austin      15.0             Robbery    fall       3122  0.507747
1          Austin      15.0             Robbery  summer       3092  0.502868
2          Austin      15.0             Robbery  winter       2898  0.471317
3  West Englewood       7.0  Aggravated Battery  summer       2659  0.432447
4          Austin      15.0  Aggravated Battery  summer       2598  0.422526


     name  district                fbicodedesc  season  frequency   percent
3007  NaN      18.0  Homicide 1st & 2nd Degree    fall          1  0.000163
3008  NaN      17.0  Homicide 1st & 2nd Degree  winter          1  0.000163
3009  NaN      17.0  Homicide 1st & 2nd Degree    fall          1  0.000163
3010  NaN       1.0  Homicide 1st & 2nd Degree  spring          1  0.000163
3011  NaN       1.0  Homicide 1st & 2nd Degree    fall          1  0.000163

That’s it for now. More R/Python-Pandas/Julia in future blogs.